OICM_CNN.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:17:54.668266: I tensorflow/core/util/port.cc:153] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.\n",
      "2024-09-02 17:17:54.875008: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:485] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered\n",
      "2024-09-02 17:17:54.957471: E external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:8454] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered\n",
      "2024-09-02 17:17:54.979142: E external/local_xla/xla/stream_executor/cuda/cuda_blas.cc:1452] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered\n",
      "2024-09-02 17:17:55.154846: I tensorflow/core/platform/cpu_feature_guard.cc:210] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.\n",
      "To enable the following instructions: AVX2 AVX512F AVX512_VNNI AVX512_BF16 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.\n",
      "2024-09-02 17:17:56.001938: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT\n",
      "/home/matt/CIFAR10_CNN_Project/venv/lib/python3.12/site-packages/keras/src/layers/convolutional/base_conv.py:107: UserWarning: Do not pass an `input_shape`/`input_dim` argument to a layer. When using Sequential models, prefer using an `Input(shape)` object as the first layer in the model instead.\n",
      "  super().__init__(activity_regularizer=activity_regularizer, **kwargs)\n",
      "WARNING: All log messages before absl::InitializeLog() is called are written to STDERR\n",
      "I0000 00:00:1725268677.660476 1015982 cuda_executor.cc:1001] could not open file to read NUMA node: /sys/bus/pci/devices/0000:01:00.0/numa_node\n",
      "Your kernel may have been built without NUMA support.\n",
      "I0000 00:00:1725268677.805952 1015982 cuda_executor.cc:1001] could not open file to read NUMA node: /sys/bus/pci/devices/0000:01:00.0/numa_node\n",
      "Your kernel may have been built without NUMA support.\n",
      "I0000 00:00:1725268677.806019 1015982 cuda_executor.cc:1001] could not open file to read NUMA node: /sys/bus/pci/devices/0000:01:00.0/numa_node\n",
      "Your kernel may have been built without NUMA support.\n",
      "I0000 00:00:1725268677.808150 1015982 cuda_executor.cc:1001] could not open file to read NUMA node: /sys/bus/pci/devices/0000:01:00.0/numa_node\n",
      "Your kernel may have been built without NUMA support.\n",
      "I0000 00:00:1725268677.808195 1015982 cuda_executor.cc:1001] could not open file to read NUMA node: /sys/bus/pci/devices/0000:01:00.0/numa_node\n",
      "Your kernel may have been built without NUMA support.\n",
      "I0000 00:00:1725268677.808216 1015982 cuda_executor.cc:1001] could not open file to read NUMA node: /sys/bus/pci/devices/0000:01:00.0/numa_node\n",
      "Your kernel may have been built without NUMA support.\n",
      "I0000 00:00:1725268677.931324 1015982 cuda_executor.cc:1001] could not open file to read NUMA node: /sys/bus/pci/devices/0000:01:00.0/numa_node\n",
      "Your kernel may have been built without NUMA support.\n",
      "I0000 00:00:1725268677.931374 1015982 cuda_executor.cc:1001] could not open file to read NUMA node: /sys/bus/pci/devices/0000:01:00.0/numa_node\n",
      "Your kernel may have been built without NUMA support.\n",
      "2024-09-02 17:17:57.931382: I tensorflow/core/common_runtime/gpu/gpu_device.cc:2112] Could not identify NUMA node of platform GPU id 0, defaulting to 0.  Your kernel may not have been built with NUMA support.\n",
      "I0000 00:00:1725268677.931415 1015982 cuda_executor.cc:1001] could not open file to read NUMA node: /sys/bus/pci/devices/0000:01:00.0/numa_node\n",
      "Your kernel may have been built without NUMA support.\n",
      "2024-09-02 17:17:57.931940: I tensorflow/core/common_runtime/gpu/gpu_device.cc:2021] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 21770 MB memory:  -> device: 0, name: NVIDIA GeForce RTX 3090, pci bus id: 0000:01:00.0, compute capability: 8.6\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\">Model: \"sequential\"</span>\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1mModel: \"sequential\"\u001b[0m\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
       "┃<span style=\"font-weight: bold\"> Layer (type)                    </span>┃<span style=\"font-weight: bold\"> Output Shape           </span>┃<span style=\"font-weight: bold\">       Param # </span>┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
       "│ conv2d_1 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Conv2D</span>)               │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">32</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">32</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">64</span>)     │         <span style=\"color: #00af00; text-decoration-color: #00af00\">1,792</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ batch_norm_1                    │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">32</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">32</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">64</span>)     │           <span style=\"color: #00af00; text-decoration-color: #00af00\">256</span> │\n",
       "│ (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">BatchNormalization</span>)            │                        │               │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ conv2d_2 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Conv2D</span>)               │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">32</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">32</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">64</span>)     │        <span style=\"color: #00af00; text-decoration-color: #00af00\">36,928</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ batch_norm_2                    │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">32</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">32</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">64</span>)     │           <span style=\"color: #00af00; text-decoration-color: #00af00\">256</span> │\n",
       "│ (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">BatchNormalization</span>)            │                        │               │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ max_pool_1 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">MaxPooling2D</span>)       │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">16</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">16</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">64</span>)     │             <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ conv2d_3 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Conv2D</span>)               │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">16</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">16</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">128</span>)    │        <span style=\"color: #00af00; text-decoration-color: #00af00\">73,856</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ batch_norm_3                    │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">16</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">16</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">128</span>)    │           <span style=\"color: #00af00; text-decoration-color: #00af00\">512</span> │\n",
       "│ (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">BatchNormalization</span>)            │                        │               │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ conv2d_4 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Conv2D</span>)               │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">16</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">16</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">128</span>)    │       <span style=\"color: #00af00; text-decoration-color: #00af00\">147,584</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ batch_norm_4                    │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">16</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">16</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">128</span>)    │           <span style=\"color: #00af00; text-decoration-color: #00af00\">512</span> │\n",
       "│ (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">BatchNormalization</span>)            │                        │               │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ max_pool_2 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">MaxPooling2D</span>)       │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">8</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">8</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">128</span>)      │             <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ conv2d_5 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Conv2D</span>)               │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">8</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">8</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">256</span>)      │       <span style=\"color: #00af00; text-decoration-color: #00af00\">295,168</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ batch_norm_5                    │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">8</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">8</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">256</span>)      │         <span style=\"color: #00af00; text-decoration-color: #00af00\">1,024</span> │\n",
       "│ (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">BatchNormalization</span>)            │                        │               │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ conv2d_6 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Conv2D</span>)               │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">8</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">8</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">256</span>)      │       <span style=\"color: #00af00; text-decoration-color: #00af00\">590,080</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ batch_norm_6                    │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">8</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">8</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">256</span>)      │         <span style=\"color: #00af00; text-decoration-color: #00af00\">1,024</span> │\n",
       "│ (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">BatchNormalization</span>)            │                        │               │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ max_pool_3 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">MaxPooling2D</span>)       │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">4</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">4</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">256</span>)      │             <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ flatten (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Flatten</span>)               │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">4096</span>)           │             <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ dense_1 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                 │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">512</span>)            │     <span style=\"color: #00af00; text-decoration-color: #00af00\">2,097,664</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ dropout_1 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dropout</span>)             │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">512</span>)            │             <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ dense_2 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                 │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">10</span>)             │         <span style=\"color: #00af00; text-decoration-color: #00af00\">5,130</span> │\n",
       "└─────────────────────────────────┴────────────────────────┴───────────────┘\n",
       "</pre>\n"
      ],
      "text/plain": [
       "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
       "┃\u001b[1m \u001b[0m\u001b[1mLayer (type)                   \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape          \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m      Param #\u001b[0m\u001b[1m \u001b[0m┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
       "│ conv2d_1 (\u001b[38;5;33mConv2D\u001b[0m)               │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m32\u001b[0m, \u001b[38;5;34m32\u001b[0m, \u001b[38;5;34m64\u001b[0m)     │         \u001b[38;5;34m1,792\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ batch_norm_1                    │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m32\u001b[0m, \u001b[38;5;34m32\u001b[0m, \u001b[38;5;34m64\u001b[0m)     │           \u001b[38;5;34m256\u001b[0m │\n",
       "│ (\u001b[38;5;33mBatchNormalization\u001b[0m)            │                        │               │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ conv2d_2 (\u001b[38;5;33mConv2D\u001b[0m)               │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m32\u001b[0m, \u001b[38;5;34m32\u001b[0m, \u001b[38;5;34m64\u001b[0m)     │        \u001b[38;5;34m36,928\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ batch_norm_2                    │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m32\u001b[0m, \u001b[38;5;34m32\u001b[0m, \u001b[38;5;34m64\u001b[0m)     │           \u001b[38;5;34m256\u001b[0m │\n",
       "│ (\u001b[38;5;33mBatchNormalization\u001b[0m)            │                        │               │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ max_pool_1 (\u001b[38;5;33mMaxPooling2D\u001b[0m)       │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m16\u001b[0m, \u001b[38;5;34m16\u001b[0m, \u001b[38;5;34m64\u001b[0m)     │             \u001b[38;5;34m0\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ conv2d_3 (\u001b[38;5;33mConv2D\u001b[0m)               │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m16\u001b[0m, \u001b[38;5;34m16\u001b[0m, \u001b[38;5;34m128\u001b[0m)    │        \u001b[38;5;34m73,856\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ batch_norm_3                    │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m16\u001b[0m, \u001b[38;5;34m16\u001b[0m, \u001b[38;5;34m128\u001b[0m)    │           \u001b[38;5;34m512\u001b[0m │\n",
       "│ (\u001b[38;5;33mBatchNormalization\u001b[0m)            │                        │               │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ conv2d_4 (\u001b[38;5;33mConv2D\u001b[0m)               │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m16\u001b[0m, \u001b[38;5;34m16\u001b[0m, \u001b[38;5;34m128\u001b[0m)    │       \u001b[38;5;34m147,584\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ batch_norm_4                    │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m16\u001b[0m, \u001b[38;5;34m16\u001b[0m, \u001b[38;5;34m128\u001b[0m)    │           \u001b[38;5;34m512\u001b[0m │\n",
       "│ (\u001b[38;5;33mBatchNormalization\u001b[0m)            │                        │               │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ max_pool_2 (\u001b[38;5;33mMaxPooling2D\u001b[0m)       │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m8\u001b[0m, \u001b[38;5;34m8\u001b[0m, \u001b[38;5;34m128\u001b[0m)      │             \u001b[38;5;34m0\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ conv2d_5 (\u001b[38;5;33mConv2D\u001b[0m)               │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m8\u001b[0m, \u001b[38;5;34m8\u001b[0m, \u001b[38;5;34m256\u001b[0m)      │       \u001b[38;5;34m295,168\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ batch_norm_5                    │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m8\u001b[0m, \u001b[38;5;34m8\u001b[0m, \u001b[38;5;34m256\u001b[0m)      │         \u001b[38;5;34m1,024\u001b[0m │\n",
       "│ (\u001b[38;5;33mBatchNormalization\u001b[0m)            │                        │               │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ conv2d_6 (\u001b[38;5;33mConv2D\u001b[0m)               │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m8\u001b[0m, \u001b[38;5;34m8\u001b[0m, \u001b[38;5;34m256\u001b[0m)      │       \u001b[38;5;34m590,080\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ batch_norm_6                    │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m8\u001b[0m, \u001b[38;5;34m8\u001b[0m, \u001b[38;5;34m256\u001b[0m)      │         \u001b[38;5;34m1,024\u001b[0m │\n",
       "│ (\u001b[38;5;33mBatchNormalization\u001b[0m)            │                        │               │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ max_pool_3 (\u001b[38;5;33mMaxPooling2D\u001b[0m)       │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m4\u001b[0m, \u001b[38;5;34m4\u001b[0m, \u001b[38;5;34m256\u001b[0m)      │             \u001b[38;5;34m0\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ flatten (\u001b[38;5;33mFlatten\u001b[0m)               │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m4096\u001b[0m)           │             \u001b[38;5;34m0\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ dense_1 (\u001b[38;5;33mDense\u001b[0m)                 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m512\u001b[0m)            │     \u001b[38;5;34m2,097,664\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ dropout_1 (\u001b[38;5;33mDropout\u001b[0m)             │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m512\u001b[0m)            │             \u001b[38;5;34m0\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ dense_2 (\u001b[38;5;33mDense\u001b[0m)                 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m10\u001b[0m)             │         \u001b[38;5;34m5,130\u001b[0m │\n",
       "└─────────────────────────────────┴────────────────────────┴───────────────┘\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Total params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">3,251,786</span> (12.40 MB)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Total params: \u001b[0m\u001b[38;5;34m3,251,786\u001b[0m (12.40 MB)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">3,249,994</span> (12.40 MB)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m3,249,994\u001b[0m (12.40 MB)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Non-trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">1,792</span> (7.00 KB)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Non-trainable params: \u001b[0m\u001b[38;5;34m1,792\u001b[0m (7.00 KB)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Learning rate for epoch 1 is 0.0001\n",
      "Epoch 1/80\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/matt/CIFAR10_CNN_Project/venv/lib/python3.12/site-packages/keras/src/trainers/data_adapters/py_dataset_adapter.py:122: UserWarning: Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored.\n",
      "  self._warn_if_super_not_called()\n",
      "WARNING: All log messages before absl::InitializeLog() is called are written to STDERR\n",
      "I0000 00:00:1725268680.777243 1016928 service.cc:146] XLA service 0x7f17fc02cd40 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:\n",
      "I0000 00:00:1725268680.777271 1016928 service.cc:154]   StreamExecutor device (0): NVIDIA GeForce RTX 3090, Compute Capability 8.6\n",
      "2024-09-02 17:18:00.852315: I tensorflow/compiler/mlir/tensorflow/utils/dump_mlir_util.cc:268] disabling MLIR crash reproducer, set env var `MLIR_CRASH_REPRODUCER_DIRECTORY` to enable.\n",
      "2024-09-02 17:18:01.087736: I external/local_xla/xla/stream_executor/cuda/cuda_dnn.cc:531] Loaded cuDNN version 8907\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m 16/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m8s\u001b[0m 11ms/step - accuracy: 0.1341 - loss: 5.1167 "
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "I0000 00:00:1725268684.028095 1016928 device_compiler.h:188] Compiled cluster using XLA!  This line is logged at most once for the lifetime of the process.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m18s\u001b[0m 17ms/step - accuracy: 0.2997 - loss: 3.0464 - val_accuracy: 0.4903 - val_loss: 2.2624 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 2 is 0.0001\n",
      "Epoch 2/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m4s\u001b[0m 6ms/step - accuracy: 0.3750 - loss: 2.6280"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:18:17.289941: I tensorflow/core/framework/local_rendezvous.cc:404] Local rendezvous is aborting with status: OUT_OF_RANGE: End of sequence\n",
      "\t [[{{node IteratorGetNext}}]]\n",
      "2024-09-02 17:18:17.289979: I tensorflow/core/framework/local_rendezvous.cc:404] Local rendezvous is aborting with status: OUT_OF_RANGE: End of sequence\n",
      "\t [[{{node IteratorGetNext}}]]\n",
      "\t [[IteratorGetNext/_4]]\n",
      "2024-09-02 17:18:17.289990: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n",
      "/home/matt/miniconda3/lib/python3.12/contextlib.py:158: UserWarning: Your input ran out of data; interrupting training. Make sure that your dataset or generator can generate at least `steps_per_epoch * epochs` batches. You may need to use the `.repeat()` function when building your dataset.\n",
      "  self.gen.throw(value)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 549us/step - accuracy: 0.3750 - loss: 2.6280 - val_accuracy: 0.4933 - val_loss: 2.2507 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 3 is 0.0001\n",
      "Epoch 3/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.4444 - loss: 2.3573 - val_accuracy: 0.5387 - val_loss: 2.0356 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 4 is 0.0001\n",
      "Epoch 4/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m4s\u001b[0m 5ms/step - accuracy: 0.4219 - loss: 2.2464"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:18:26.772228: I tensorflow/core/framework/local_rendezvous.cc:404] Local rendezvous is aborting with status: OUT_OF_RANGE: End of sequence\n",
      "\t [[{{node IteratorGetNext}}]]\n",
      "\t [[IteratorGetNext/_4]]\n",
      "2024-09-02 17:18:26.772271: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 554us/step - accuracy: 0.4219 - loss: 2.2464 - val_accuracy: 0.5348 - val_loss: 2.0449 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 5 is 0.0001\n",
      "Epoch 5/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.5104 - loss: 2.0763 - val_accuracy: 0.5917 - val_loss: 1.7881 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 6 is 0.0001\n",
      "Epoch 6/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m4s\u001b[0m 5ms/step - accuracy: 0.5312 - loss: 1.7886"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:18:36.121873: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 525us/step - accuracy: 0.5312 - loss: 1.7886 - val_accuracy: 0.5869 - val_loss: 1.7970 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 7 is 0.0001\n",
      "Epoch 7/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.5447 - loss: 1.8785 - val_accuracy: 0.6149 - val_loss: 1.6005 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 8 is 0.0001\n",
      "Epoch 8/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m3s\u001b[0m 5ms/step - accuracy: 0.5781 - loss: 1.5664"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:18:45.290029: I tensorflow/core/framework/local_rendezvous.cc:404] Local rendezvous is aborting with status: OUT_OF_RANGE: End of sequence\n",
      "\t [[{{node IteratorGetNext}}]]\n",
      "\t [[IteratorGetNext/_4]]\n",
      "2024-09-02 17:18:45.290057: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 524us/step - accuracy: 0.5781 - loss: 1.5664 - val_accuracy: 0.6172 - val_loss: 1.5991 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 9 is 0.0001\n",
      "Epoch 9/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.5843 - loss: 1.6813 - val_accuracy: 0.6527 - val_loss: 1.4174 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 10 is 0.0001\n",
      "Epoch 10/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m4s\u001b[0m 5ms/step - accuracy: 0.5938 - loss: 1.4148"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:18:54.318548: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 525us/step - accuracy: 0.5938 - loss: 1.4148 - val_accuracy: 0.6552 - val_loss: 1.4166 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 11 is 0.0001\n",
      "Epoch 11/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.6091 - loss: 1.5276 - val_accuracy: 0.6975 - val_loss: 1.2385 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 12 is 0.0001\n",
      "Epoch 12/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m3s\u001b[0m 5ms/step - accuracy: 0.5781 - loss: 1.4611"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:19:03.357272: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 513us/step - accuracy: 0.5781 - loss: 1.4611 - val_accuracy: 0.6940 - val_loss: 1.2429 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 13 is 0.0001\n",
      "Epoch 13/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.6366 - loss: 1.3936 - val_accuracy: 0.6934 - val_loss: 1.2053 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 14 is 0.0001\n",
      "Epoch 14/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m3s\u001b[0m 5ms/step - accuracy: 0.7031 - loss: 1.3279"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:19:12.347784: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 515us/step - accuracy: 0.7031 - loss: 1.3279 - val_accuracy: 0.6923 - val_loss: 1.1988 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 15 is 0.0001\n",
      "Epoch 15/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.6570 - loss: 1.2912 - val_accuracy: 0.7038 - val_loss: 1.1154 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 16 is 0.0001\n",
      "Epoch 16/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m4s\u001b[0m 5ms/step - accuracy: 0.6406 - loss: 1.2862"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:19:21.317417: I tensorflow/core/framework/local_rendezvous.cc:404] Local rendezvous is aborting with status: OUT_OF_RANGE: End of sequence\n",
      "\t [[{{node IteratorGetNext}}]]\n",
      "\t [[IteratorGetNext/_4]]\n",
      "2024-09-02 17:19:21.317445: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 508us/step - accuracy: 0.6406 - loss: 1.2862 - val_accuracy: 0.7024 - val_loss: 1.1177 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 17 is 0.0001\n",
      "Epoch 17/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.6787 - loss: 1.1971 - val_accuracy: 0.7224 - val_loss: 1.0388 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 18 is 0.0001\n",
      "Epoch 18/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m4s\u001b[0m 5ms/step - accuracy: 0.7031 - loss: 1.0506"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:19:30.306268: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 509us/step - accuracy: 0.7031 - loss: 1.0506 - val_accuracy: 0.7188 - val_loss: 1.0498 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 19 is 0.0001\n",
      "Epoch 19/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.6972 - loss: 1.1184 - val_accuracy: 0.7232 - val_loss: 1.0338 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 20 is 0.0001\n",
      "Epoch 20/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m4s\u001b[0m 5ms/step - accuracy: 0.7188 - loss: 0.9331"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:19:39.262969: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 514us/step - accuracy: 0.7188 - loss: 0.9331 - val_accuracy: 0.7235 - val_loss: 1.0411 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 21 is 0.0001\n",
      "Epoch 21/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.7097 - loss: 1.0546 - val_accuracy: 0.7426 - val_loss: 0.9550 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 22 is 0.0001\n",
      "Epoch 22/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 524us/step - accuracy: 0.7031 - loss: 0.9891 - val_accuracy: 0.7445 - val_loss: 0.9526 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 23 is 0.0001\n",
      "Epoch 23/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 11ms/step - accuracy: 0.7247 - loss: 1.0018 - val_accuracy: 0.7534 - val_loss: 0.8928 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 24 is 0.0001\n",
      "Epoch 24/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m4s\u001b[0m 5ms/step - accuracy: 0.8594 - loss: 0.7437"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:19:57.210062: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 530us/step - accuracy: 0.8594 - loss: 0.7437 - val_accuracy: 0.7544 - val_loss: 0.8911 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 25 is 0.0001\n",
      "Epoch 25/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.7312 - loss: 0.9583 - val_accuracy: 0.7718 - val_loss: 0.8651 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 26 is 0.0001\n",
      "Epoch 26/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m4s\u001b[0m 5ms/step - accuracy: 0.7969 - loss: 0.7664"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:20:06.220166: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 521us/step - accuracy: 0.7969 - loss: 0.7664 - val_accuracy: 0.7711 - val_loss: 0.8626 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 27 is 0.0001\n",
      "Epoch 27/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.7504 - loss: 0.9072 - val_accuracy: 0.7794 - val_loss: 0.8200 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 28 is 0.0001\n",
      "Epoch 28/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 514us/step - accuracy: 0.6875 - loss: 1.0095 - val_accuracy: 0.7766 - val_loss: 0.8249 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 29 is 0.0001\n",
      "Epoch 29/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.7616 - loss: 0.8657 - val_accuracy: 0.7894 - val_loss: 0.7747 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 30 is 0.0001\n",
      "Epoch 30/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m4s\u001b[0m 5ms/step - accuracy: 0.7656 - loss: 0.7273"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:20:24.184946: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 525us/step - accuracy: 0.7656 - loss: 0.7273 - val_accuracy: 0.7895 - val_loss: 0.7703 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 31 is 0.0001\n",
      "Epoch 31/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.7727 - loss: 0.8166 - val_accuracy: 0.8083 - val_loss: 0.7209 - learning_rate: 1.0000e-04\n",
      "Learning rate for epoch 32 is 5e-05\n",
      "Epoch 32/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m4s\u001b[0m 5ms/step - accuracy: 0.7812 - loss: 0.8811"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:20:33.142579: I tensorflow/core/framework/local_rendezvous.cc:404] Local rendezvous is aborting with status: OUT_OF_RANGE: End of sequence\n",
      "\t [[{{node IteratorGetNext}}]]\n",
      "\t [[IteratorGetNext/_4]]\n",
      "2024-09-02 17:20:33.142620: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 519us/step - accuracy: 0.7812 - loss: 0.8811 - val_accuracy: 0.8074 - val_loss: 0.7226 - learning_rate: 5.0000e-05\n",
      "Learning rate for epoch 33 is 5e-05\n",
      "Epoch 33/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.7906 - loss: 0.7561 - val_accuracy: 0.8146 - val_loss: 0.6844 - learning_rate: 5.0000e-05\n",
      "Learning rate for epoch 34 is 5e-05\n",
      "Epoch 34/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m4s\u001b[0m 5ms/step - accuracy: 0.7656 - loss: 0.8559"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:20:42.089733: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 519us/step - accuracy: 0.7656 - loss: 0.8559 - val_accuracy: 0.8155 - val_loss: 0.6863 - learning_rate: 5.0000e-05\n",
      "Learning rate for epoch 35 is 5e-05\n",
      "Epoch 35/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.7979 - loss: 0.7284 - val_accuracy: 0.8123 - val_loss: 0.6969 - learning_rate: 5.0000e-05\n",
      "Learning rate for epoch 36 is 5e-05\n",
      "Epoch 36/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m3s\u001b[0m 5ms/step - accuracy: 0.7969 - loss: 0.7265"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:20:51.166622: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 513us/step - accuracy: 0.7969 - loss: 0.7265 - val_accuracy: 0.8139 - val_loss: 0.6949 - learning_rate: 5.0000e-05\n",
      "Learning rate for epoch 37 is 5e-05\n",
      "Epoch 37/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.8054 - loss: 0.7028 - val_accuracy: 0.8228 - val_loss: 0.6517 - learning_rate: 5.0000e-05\n",
      "Learning rate for epoch 38 is 5e-05\n",
      "Epoch 38/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 520us/step - accuracy: 0.8281 - loss: 0.5602 - val_accuracy: 0.8216 - val_loss: 0.6544 - learning_rate: 5.0000e-05\n",
      "Learning rate for epoch 39 is 5e-05\n",
      "Epoch 39/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.8072 - loss: 0.6902 - val_accuracy: 0.8240 - val_loss: 0.6411 - learning_rate: 5.0000e-05\n",
      "Learning rate for epoch 40 is 5e-05\n",
      "Epoch 40/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m3s\u001b[0m 5ms/step - accuracy: 0.7500 - loss: 0.8191"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:21:09.101715: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 519us/step - accuracy: 0.7500 - loss: 0.8191 - val_accuracy: 0.8246 - val_loss: 0.6395 - learning_rate: 5.0000e-05\n",
      "Learning rate for epoch 41 is 5e-05\n",
      "Epoch 41/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.8128 - loss: 0.6717 - val_accuracy: 0.8271 - val_loss: 0.6287 - learning_rate: 5.0000e-05\n",
      "Learning rate for epoch 42 is 5e-05\n",
      "Epoch 42/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m4s\u001b[0m 5ms/step - accuracy: 0.7969 - loss: 0.7193"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:21:18.029158: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 511us/step - accuracy: 0.7969 - loss: 0.7193 - val_accuracy: 0.8264 - val_loss: 0.6300 - learning_rate: 5.0000e-05\n",
      "Learning rate for epoch 43 is 5e-05\n",
      "Epoch 43/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.8180 - loss: 0.6452 - val_accuracy: 0.8424 - val_loss: 0.5773 - learning_rate: 5.0000e-05\n",
      "Learning rate for epoch 44 is 5e-05\n",
      "Epoch 44/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m3s\u001b[0m 5ms/step - accuracy: 0.8438 - loss: 0.5875"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:21:27.047816: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 521us/step - accuracy: 0.8438 - loss: 0.5875 - val_accuracy: 0.8425 - val_loss: 0.5777 - learning_rate: 5.0000e-05\n",
      "Learning rate for epoch 45 is 5e-05\n",
      "Epoch 45/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.8197 - loss: 0.6336 - val_accuracy: 0.8304 - val_loss: 0.6115 - learning_rate: 5.0000e-05\n",
      "Learning rate for epoch 46 is 5e-05\n",
      "Epoch 46/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m3s\u001b[0m 5ms/step - accuracy: 0.7969 - loss: 0.6344"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:21:36.036217: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 514us/step - accuracy: 0.7969 - loss: 0.6344 - val_accuracy: 0.8292 - val_loss: 0.6190 - learning_rate: 5.0000e-05\n",
      "Learning rate for epoch 47 is 5e-05\n",
      "Epoch 47/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 11ms/step - accuracy: 0.8253 - loss: 0.6150 - val_accuracy: 0.8319 - val_loss: 0.6113 - learning_rate: 5.0000e-05\n",
      "Learning rate for epoch 48 is 5e-05\n",
      "Epoch 48/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m3s\u001b[0m 5ms/step - accuracy: 0.8281 - loss: 0.5564\n",
      "Epoch 48: ReduceLROnPlateau reducing learning rate to 2.499999936844688e-05.\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 517us/step - accuracy: 0.8281 - loss: 0.5564 - val_accuracy: 0.8318 - val_loss: 0.6097 - learning_rate: 5.0000e-05\n",
      "Learning rate for epoch 49 is 5e-05\n",
      "Epoch 49/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.8269 - loss: 0.6061 - val_accuracy: 0.8430 - val_loss: 0.5768 - learning_rate: 5.0000e-05\n",
      "Learning rate for epoch 50 is 5e-05\n",
      "Epoch 50/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 510us/step - accuracy: 0.7812 - loss: 0.7201 - val_accuracy: 0.8419 - val_loss: 0.5778 - learning_rate: 5.0000e-05\n",
      "Learning rate for epoch 51 is 5e-05\n",
      "Epoch 51/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.8335 - loss: 0.5887 - val_accuracy: 0.8495 - val_loss: 0.5543 - learning_rate: 5.0000e-05\n",
      "Learning rate for epoch 52 is 1e-05\n",
      "Epoch 52/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m4s\u001b[0m 5ms/step - accuracy: 0.8281 - loss: 0.5203"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:22:02.961059: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 518us/step - accuracy: 0.8281 - loss: 0.5203 - val_accuracy: 0.8492 - val_loss: 0.5540 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 53 is 1e-05\n",
      "Epoch 53/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.8411 - loss: 0.5657 - val_accuracy: 0.8526 - val_loss: 0.5293 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 54 is 1e-05\n",
      "Epoch 54/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m4s\u001b[0m 5ms/step - accuracy: 0.7656 - loss: 0.6163"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:22:12.076191: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 521us/step - accuracy: 0.7656 - loss: 0.6163 - val_accuracy: 0.8525 - val_loss: 0.5291 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 55 is 1e-05\n",
      "Epoch 55/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.8453 - loss: 0.5435 - val_accuracy: 0.8543 - val_loss: 0.5319 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 56 is 1e-05\n",
      "Epoch 56/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 512us/step - accuracy: 0.9531 - loss: 0.3247 - val_accuracy: 0.8543 - val_loss: 0.5318 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 57 is 1e-05\n",
      "Epoch 57/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.8488 - loss: 0.5388 - val_accuracy: 0.8520 - val_loss: 0.5394 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 58 is 1e-05\n",
      "Epoch 58/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m3s\u001b[0m 5ms/step - accuracy: 0.9219 - loss: 0.3423"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:22:30.168273: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 7338431921674375828\n",
      "2024-09-02 17:22:30.168312: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 507us/step - accuracy: 0.9219 - loss: 0.3423 - val_accuracy: 0.8523 - val_loss: 0.5395 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 59 is 1e-05\n",
      "Epoch 59/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.8497 - loss: 0.5332 - val_accuracy: 0.8568 - val_loss: 0.5207 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 60 is 1e-05\n",
      "Epoch 60/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m4s\u001b[0m 5ms/step - accuracy: 0.8750 - loss: 0.4386"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:22:39.221483: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 526us/step - accuracy: 0.8750 - loss: 0.4386 - val_accuracy: 0.8576 - val_loss: 0.5201 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 61 is 1e-05\n",
      "Epoch 61/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.8524 - loss: 0.5270 - val_accuracy: 0.8524 - val_loss: 0.5309 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 62 is 1e-05\n",
      "Epoch 62/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m3s\u001b[0m 5ms/step - accuracy: 0.8438 - loss: 0.4579"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:22:48.314365: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 518us/step - accuracy: 0.8438 - loss: 0.4579 - val_accuracy: 0.8524 - val_loss: 0.5305 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 63 is 1e-05\n",
      "Epoch 63/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.8523 - loss: 0.5226 - val_accuracy: 0.8552 - val_loss: 0.5220 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 64 is 1e-05\n",
      "Epoch 64/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m3s\u001b[0m 5ms/step - accuracy: 0.8750 - loss: 0.4568"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:22:57.335231: I tensorflow/core/framework/local_rendezvous.cc:404] Local rendezvous is aborting with status: OUT_OF_RANGE: End of sequence\n",
      "\t [[{{node IteratorGetNext}}]]\n",
      "\t [[IteratorGetNext/_4]]\n",
      "2024-09-02 17:22:57.335281: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 523us/step - accuracy: 0.8750 - loss: 0.4568 - val_accuracy: 0.8555 - val_loss: 0.5219 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 65 is 1e-05\n",
      "Epoch 65/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.8543 - loss: 0.5207 - val_accuracy: 0.8605 - val_loss: 0.5082 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 66 is 1e-05\n",
      "Epoch 66/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m4s\u001b[0m 5ms/step - accuracy: 0.8438 - loss: 0.6187"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:23:06.393971: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 513us/step - accuracy: 0.8438 - loss: 0.6187 - val_accuracy: 0.8605 - val_loss: 0.5086 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 67 is 1e-05\n",
      "Epoch 67/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.8543 - loss: 0.5162 - val_accuracy: 0.8593 - val_loss: 0.5150 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 68 is 1e-05\n",
      "Epoch 68/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m3s\u001b[0m 5ms/step - accuracy: 0.8594 - loss: 0.4554"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:23:15.479595: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 515us/step - accuracy: 0.8594 - loss: 0.4554 - val_accuracy: 0.8594 - val_loss: 0.5153 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 69 is 1e-05\n",
      "Epoch 69/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.8577 - loss: 0.5100 - val_accuracy: 0.8587 - val_loss: 0.5114 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 70 is 1e-05\n",
      "Epoch 70/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m3s\u001b[0m 5ms/step - accuracy: 0.8906 - loss: 0.4385"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:23:24.399811: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Epoch 70: ReduceLROnPlateau reducing learning rate to 4.999999873689376e-06.\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 519us/step - accuracy: 0.8906 - loss: 0.4385 - val_accuracy: 0.8588 - val_loss: 0.5114 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 71 is 1e-05\n",
      "Epoch 71/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.8611 - loss: 0.4964 - val_accuracy: 0.8613 - val_loss: 0.5042 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 72 is 1e-05\n",
      "Epoch 72/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m4s\u001b[0m 6ms/step - accuracy: 0.8906 - loss: 0.3925"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:23:33.426858: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 532us/step - accuracy: 0.8906 - loss: 0.3925 - val_accuracy: 0.8609 - val_loss: 0.5040 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 73 is 1e-05\n",
      "Epoch 73/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.8585 - loss: 0.4987 - val_accuracy: 0.8648 - val_loss: 0.4972 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 74 is 1e-05\n",
      "Epoch 74/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 527us/step - accuracy: 0.9062 - loss: 0.3515 - val_accuracy: 0.8654 - val_loss: 0.4969 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 75 is 1e-05\n",
      "Epoch 75/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.8607 - loss: 0.4953 - val_accuracy: 0.8679 - val_loss: 0.4849 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 76 is 1e-05\n",
      "Epoch 76/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m4s\u001b[0m 5ms/step - accuracy: 0.8594 - loss: 0.4814"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:23:51.399762: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 523us/step - accuracy: 0.8594 - loss: 0.4814 - val_accuracy: 0.8676 - val_loss: 0.4849 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 77 is 1e-05\n",
      "Epoch 77/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.8619 - loss: 0.4858 - val_accuracy: 0.8612 - val_loss: 0.5146 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 78 is 1e-05\n",
      "Epoch 78/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m3s\u001b[0m 5ms/step - accuracy: 0.8125 - loss: 0.5066"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:24:00.450300: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 515us/step - accuracy: 0.8125 - loss: 0.5066 - val_accuracy: 0.8609 - val_loss: 0.5146 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 79 is 1e-05\n",
      "Epoch 79/80\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 11ms/step - accuracy: 0.8629 - loss: 0.4860 - val_accuracy: 0.8665 - val_loss: 0.4916 - learning_rate: 1.0000e-05\n",
      "Learning rate for epoch 80 is 1e-05\n",
      "Epoch 80/80\n",
      "\u001b[1m  1/781\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m3s\u001b[0m 5ms/step - accuracy: 0.8438 - loss: 0.5229"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:24:09.451889: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Epoch 80: ReduceLROnPlateau reducing learning rate to 4.999999873689376e-06.\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 522us/step - accuracy: 0.8438 - loss: 0.5229 - val_accuracy: 0.8663 - val_loss: 0.4919 - learning_rate: 1.0000e-05\n",
      "\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 345ms/step\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA+kAAAGJCAYAAAD2VnIMAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjkuMiwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8hTgPZAAAACXBIWXMAAA9hAAAPYQGoP6dpAADUSElEQVR4nOzdd3xT9frA8U+Spuneu7SUlr33VMArCKIIqCigooh6VXDh9ee8rqvi9Tq47onoFRRBxT3Ye8iWVSiUDrr3btrk/P44SdrSmdLS9bxfr7ySnpxz8k3Fnjx5vt/n0SiKoiCEEEIIIYQQQogWp23pAQghhBBCCCGEEEIlQboQQgghhBBCCNFKSJAuhBBCCCGEEEK0EhKkCyGEEEIIIYQQrYQE6UIIIYQQQgghRCshQboQQgghhBBCCNFKSJAuhBBCCCGEEEK0EhKkCyGEEEIIIYQQrYQE6UIIIYQQQgghRCshQboQbZRGo+HZZ5+1+7izZ8+i0WhYtmxZk49JCCGEEK2XfHYQom2QIF2IC7Bs2TI0Gg0ajYZt27ZVe15RFMLCwtBoNFx99dUtMMKm8csvv6DRaAgJCcFsNrf0cIQQQog2qz1/dti0aRMajYbVq1e39FCEaNMkSBeiCTg5ObFixYpq2zdv3kxiYiIGg6EFRtV0li9fTkREBMnJyWzYsKGlhyOEEEK0ee39s4MQovEkSBeiCUyZMoVVq1ZRXl5eZfuKFSsYMmQIQUFBLTSyC1dYWMj333/PokWLGDRoEMuXL2/pIdWqsLCwpYcghBBCNEh7/uwghLgwEqQL0QRmz55NZmYma9eutW0zGo2sXr2aOXPm1HhMYWEhDz/8MGFhYRgMBnr06MGrr76KoihV9istLeWhhx7C398fd3d3rrnmGhITE2s857lz57j99tsJDAzEYDDQp08fli5dekHv7bvvvqO4uJiZM2cya9Ysvv32W0pKSqrtV1JSwrPPPkv37t1xcnIiODiYa6+9ltOnT9v2MZvN/Pe//6Vfv344OTnh7+/P5MmT2bt3L1D3mrfz19E9++yzaDQajh07xpw5c/D29uaSSy4B4PDhw9x2221ERkbi5OREUFAQt99+O5mZmTX+zubPn09ISAgGg4EuXbpwzz33YDQaOXPmDBqNhjfeeKPacTt27ECj0fDll1/a+ysVQggh2vVnh/qcOXOGmTNn4uPjg4uLCyNHjuTnn3+utt9bb71Fnz59cHFxwdvbm6FDh1aZfZCfn8+DDz5IREQEBoOBgIAAJk6cyP79+5t1/EI0N4eWHoAQ7UFERASjRo3iyy+/5MorrwTg119/JTc3l1mzZvHmm29W2V9RFK655ho2btzI/PnzGThwIL///juPPPII586dqxIU3nHHHXzxxRfMmTOH0aNHs2HDBq666qpqY0hNTWXkyJFoNBoWLlyIv78/v/76K/PnzycvL48HH3ywUe9t+fLlXHbZZQQFBTFr1iwee+wxfvzxR2bOnGnbx2QycfXVV7N+/XpmzZrFAw88QH5+PmvXruXIkSNERUUBMH/+fJYtW8aVV17JHXfcQXl5OVu3bmXXrl0MHTq0UeObOXMm3bp146WXXrJ9SFm7di1nzpxh3rx5BAUFcfToUT788EOOHj3Krl270Gg0ACQlJTF8+HBycnK466676NmzJ+fOnWP16tUUFRURGRnJmDFjWL58OQ899FC134u7uzvTpk1r1LiFEEJ0bO35s0NdUlNTGT16NEVFRdx///34+vry2Wefcc0117B69WpmzJgBwEcffcT999/P9ddfzwMPPEBJSQmHDx9m9+7dti8x7r77blavXs3ChQvp3bs3mZmZbNu2jePHjzN48OAmH7sQF40ihGi0Tz/9VAGUP//8U3n77bcVd3d3paioSFEURZk5c6Zy2WWXKYqiKJ07d1auuuoq23Fr1qxRAOWFF16ocr7rr79e0Wg0SkxMjKIoinLw4EEFUO69994q+82ZM0cBlGeeeca2bf78+UpwcLCSkZFRZd9Zs2Ypnp6etnHFxsYqgPLpp5/W+/5SU1MVBwcH5aOPPrJtGz16tDJt2rQq+y1dulQBlNdff73aOcxms6IoirJhwwYFUO6///5a96lrbOe/32eeeUYBlNmzZ1fb1/peK/vyyy8VQNmyZYtt29y5cxWtVqv8+eeftY7pgw8+UADl+PHjtueMRqPi5+en3HrrrdWOE0IIIerSnj87bNy4UQGUVatW1brPgw8+qADK1q1bbdvy8/OVLl26KBEREYrJZFIURVGmTZum9OnTp87X8/T0VBYsWFDnPkK0RTLdXYgmcsMNN1BcXMxPP/1Efn4+P/30U63T1X755Rd0Oh33339/le0PP/wwiqLw66+/2vYDqu13/jfbiqLwzTffMHXqVBRFISMjw3abNGkSubm5jZr69dVXX6HVarnuuuts22bPns2vv/5Kdna2bds333yDn58f9913X7VzWLPW33zzDRqNhmeeeabWfRrj7rvvrrbN2dnZ9rikpISMjAxGjhwJYPs9mM1m1qxZw9SpU2vM4lvHdMMNN+Dk5FRlLf7vv/9ORkYGN998c6PHLYQQQrTHzw71+eWXXxg+fLhtiRqAm5sbd911F2fPnuXYsWMAeHl5kZiYyJ9//lnruby8vNi9ezdJSUlNPk4hWpIE6UI0EX9/fyZMmMCKFSv49ttvMZlMXH/99TXuGxcXR0hICO7u7lW29+rVy/a89V6r1dqmi1v16NGjys/p6enk5OTw4Ycf4u/vX+U2b948ANLS0ux+T1988QXDhw8nMzOTmJgYYmJiGDRoEEajkVWrVtn2O336ND169MDBofYVNKdPnyYkJAQfHx+7x1GXLl26VNuWlZXFAw88QGBgIM7Ozvj7+9v2y83NBdTfWV5eHn379q3z/F5eXkydOrXKGrjly5cTGhrK3/72tyZ8J0IIITqa9vjZoT5xcXHVxlLT+3j00Udxc3Nj+PDhdOvWjQULFrB9+/Yqx7zyyiscOXKEsLAwhg8fzrPPPsuZM2eafMxCXGyyJl2IJjRnzhzuvPNOUlJSuPLKK/Hy8roor2vtXX7zzTdz66231rhP//797TrnqVOnbN9ed+vWrdrzy5cv56677rJzpHWrLaNuMplqPaZy1tzqhhtuYMeOHTzyyCMMHDgQNzc3zGYzkydPblSf97lz57Jq1Sp27NhBv379+OGHH7j33nvRauV7TiGEEBemPX12aEq9evUiOjqan376id9++41vvvmGd999l6effprnnnsOUK/3l156Kd999x1//PEH//nPf/j3v//Nt99+a1vnL0RbJEG6EE1oxowZ/P3vf2fXrl2sXLmy1v06d+7MunXryM/Pr/KN+IkTJ2zPW+/NZrMtU20VHR1d5XzW6q0mk4kJEyY0yXtZvnw5er2e//3vf+h0uirPbdu2jTfffJP4+HjCw8OJiopi9+7dlJWVodfrazxfVFQUv//+O1lZWbVm0729vQHIycmpst36rXpDZGdns379ep577jmefvpp2/ZTp05V2c/f3x8PDw+OHDlS7zknT56Mv78/y5cvZ8SIERQVFXHLLbc0eExCCCFEbdrTZ4eG6Ny5c7WxQPX3AeDq6sqNN97IjTfeiNFo5Nprr+XFF1/k8ccfx8nJCYDg4GDuvfde7r33XtLS0hg8eDAvvviiBOmiTZM0kBBNyM3Njffee49nn32WqVOn1rrflClTMJlMvP3221W2v/HGG2g0GtuFxXp/foXXJUuWVPlZp9Nx3XXX8c0339QYdKanp9v9XpYvX86ll17KjTfeyPXXX1/l9sgjjwDY2o9dd911ZGRkVHs/gK3i+nXXXYeiKLZvv2vax8PDAz8/P7Zs2VLl+XfffbfB47Z+oaCc147m/N+ZVqtl+vTp/Pjjj7YWcDWNCcDBwYHZs2fz9ddfs2zZMvr169ei2QUhhBDtR3v67NAQU6ZMYc+ePezcudO2rbCwkA8//JCIiAh69+4NUK1tqqOjI71790ZRFMrKyjCZTLYlbFYBAQGEhIRQWlraLGMX4mKRTLoQTay2KWOVTZ06lcsuu4wnn3ySs2fPMmDAAP744w++//57HnzwQds6soEDBzJ79mzeffddcnNzGT16NOvXrycmJqbaOV9++WU2btzIiBEjuPPOO+nduzdZWVns37+fdevWkZWV1eD3sHv3bmJiYli4cGGNz4eGhjJ48GCWL1/Oo48+yty5c/n8889ZtGgRe/bs4dJLL6WwsJB169Zx7733Mm3aNC677DJuueUW3nzzTU6dOmWber5161Yuu+wy22vdcccdvPzyy9xxxx0MHTqULVu2cPLkyQaP3cPDg7Fjx/LKK69QVlZGaGgof/zxB7GxsdX2femll/jjjz8YN24cd911F7169SI5OZlVq1axbdu2KlMO586dy5tvvsnGjRv597//3eDxCCGEEPVpD58dKvvmm29smfHz3+djjz1mazt3//334+Pjw2effUZsbCzffPONbSnZFVdcQVBQEGPGjCEwMJDjx4/z9ttvc9VVV+Hu7k5OTg6dOnXi+uuvZ8CAAbi5ubFu3Tr+/PNPXnvttUaNW4hWo2WKygvRPlRuo1KX89uoKIrabuShhx5SQkJCFL1er3Tr1k35z3/+Y2v9ZVVcXKzcf//9iq+vr+Lq6qpMnTpVSUhIqNZGRVHUlmkLFixQwsLCFL1erwQFBSmXX3658uGHH9r2aUgblfvuu08BlNOnT9e6z7PPPqsAyqFDhxRFUduePfnkk0qXLl1sr3399ddXOUd5ebnyn//8R+nZs6fi6Oio+Pv7K1deeaWyb98+2z5FRUXK/PnzFU9PT8Xd3V254YYblLS0tFpbsKWnp1cbW2JiojJjxgzFy8tL8fT0VGbOnKkkJSXV+DuLi4tT5s6dq/j7+ysGg0GJjIxUFixYoJSWllY7b58+fRStVqskJibW+nsRQggh6tJePzsoSkULttpu1rZrp0+fVq6//nrFy8tLcXJyUoYPH6789NNPVc71wQcfKGPHjlV8fX0Vg8GgREVFKY888oiSm5urKIqilJaWKo888ogyYMAAxd3dXXF1dVUGDBigvPvuu3WOUYi2QKMo580JFUIIUaNBgwbh4+PD+vXrW3ooQgghhBCinZI16UII0QB79+7l4MGDzJ07t6WHIoQQQggh2jHJpAshRB2OHDnCvn37eO2118jIyODMmTO2irJCCCGEEEI0NcmkCyFEHVavXs28efMoKyvjyy+/lABdCCGEEEI0K8mkCyGEEEIIIYQQrYRk0oUQQgghhBBCiFZCgnQhhBBCCCGEEKKVcGjpAVxsZrOZpKQk3N3d0Wg0LT0cIYQQAkVRyM/PJyQkBK1Wvj9vCnK9F0II0ZrYc63vcEF6UlISYWFhLT0MIYQQopqEhAQ6derU0sNoF+R6L4QQojVqyLW+wwXp7u7ugPrL8fDwaOHRCCGEEJCXl0dYWJjtGiUunFzvhRBCtCb2XOs7XJBunfLm4eEhF20hhBCtikzLbjpyvRdCCNEaNeRaLwvfhBBCCCGEEEKIVkKCdCGEEEIIIYQQopWQIF0IIYQQQgghhGglOtya9IZQFIXy8nJMJlNLD6XN0ul0ODg4yPpKIYQQrZJc6y+cXOuFEKJ5SJB+HqPRSHJyMkVFRS09lDbPxcWF4OBgHB0dW3ooQgghhI1c65uOXOuFEKLpSZBeidlsJjY2Fp1OR0hICI6OjvLtcCMoioLRaCQ9PZ3Y2Fi6deuGVisrK4QQQrQ8udY3DbnWCyFE85EgvRKj0YjZbCYsLAwXF5eWHk6b5uzsjF6vJy4uDqPRiJOTU0sPSQghhJBrfROSa70QQjQP+cqzBvJNcNOQ36MQQojWSq5RTUN+j0II0fTkL6sQQgghhBBCCNFKSJAuhBAXQWJ2EbEZhS09DCGEnRRFIbfYSEFJeUsPRQghRAchQbqoVUREBEuWLGnpYQjR5pnNCjPe3cHVb26lyCgf9IVoSzILjMRlFpGUW4yiKC09nCYn13ohhGh9JEhvBzQaTZ23Z599tlHn/fPPP7nrrruadrBCdEDZRUbS80spNJo4l13c0sMRQtjBy0WPTquhpMxEbnFZi41DrvVCCNFxSHX3diA5Odn2eOXKlTz99NNER0fbtrm5udkeK4qCyWTCwaH+//T+/v5NO1AhOqi0/NIqj7sFurfgaIQQ9nDQafF3M5CSV0JqXgkeznq0LdCyTa71QgjRcUgmvR6KolBkLG+RW0On1QUFBdlunp6eaDQa288nTpzA3d2dX3/9lSFDhmAwGNi2bRunT59m2rRpBAYG4ubmxrBhw1i3bl2V854/BU6j0fDxxx8zY8YMXFxc6NatGz/88ENT/rqFaJcqB+mpeSUtOBIhRE3qu9Y7O+ooNynkFpeRnFMs13ohhBDNSjLp9SguM9H76d9b5LWPPT8JF8em+U/02GOP8eqrrxIZGYm3tzcJCQlMmTKFF198EYPBwOeff87UqVOJjo4mPDy81vM899xzvPLKK/znP//hrbfe4qabbiIuLg4fH58mGacQ7VH6eZl0IUTrItf6quRaL4QQLUsy6R3E888/z8SJE4mKisLHx4cBAwbw97//nb59+9KtWzf+9a9/ERUVVe+35bfddhuzZ8+ma9euvPTSSxQUFLBnz56L9C6EaJvS8iuy52l5EqQLIZqHXOuFEKJ9kEx6PZz1Oo49P6nFXrupDB06tMrPBQUFPPvss/z8888kJydTXl5OcXEx8fHxdZ6nf//+tseurq54eHiQlpbWZOMUoj2qnElPzZfp7kK0Ng291mcXGjmXU4yDVku3QDd02gtfmy7XeiGEEOeTIL0eGo2myaahtSRXV9cqP//jH/9g7dq1vPrqq3Tt2hVnZ2euv/56jEZjnefR6/VVftZoNJjN5iYfrxDtSeUp7umSSRei1Wnotd5Zr6Og1ERpuYkio4lAD6eLMLqGk2u9EEK0D20/+hSNsn37dm677TZmzJgBqN+2nz17tmUHJUQ7VTkwT5NMuhBtlkajIdDDQHxWERn5pfi6OuKga70rB+VaL4QQbVPrvbKIZtWtWze+/fZbDh48yKFDh5gzZ458Sy5EM0kvqFzdvbTB1ZyFEK2Pp7MeJ70Ok6KQUVB3RrqlybVeCCHaJgnSO6jXX38db29vRo8ezdSpU5k0aRKDBw9u6WEJ0S6lVWq7VlxmoqC0vAVHI4S4EBqNBj83RwCKjK37/2W51gshRNukUTpYSicvLw9PT09yc3Px8PCo8lxJSQmxsbF06dIFJ6fWtc6sLZLfpxBQWFpOn2fU1k6OOi1Gk5n1D48jyt+thUcmWpO6rk2icWr7nTbFtamgtJwz6QU4OmjpGdSx/3vJtV4IIRrGnmu9ZNKFEKIZWSu7uzjq6OTtDEgbNiHaOkedWtW9zKTI8hUhhBBNToJ0IYRoRtb16P7uBgI8DIAUjxOirdPrtGjQoCgK5SYJ0oUQQjQtCdKFEKIZWbPmAe4GAtydqmw7n8ms8N6m0+yLy75o4xNC2E+j0aC3ZNONJinEJoQQomlJkC6EEM3ImjX3dzcQ4F53Jn17TAb//u0ET605ctHGJ4RoHL2l9VqZBOlCCCGamATpQgjRjKxr0gPcnQj0UDPpqbVk0k+nFwBwJr0As1mm0ArRmjk6qB+hJJMuhBCiqUmQLoQQzSgtv+Fr0s9mFAJQWm4mJU/WrQvRmtky6eUSpAshhGhaEqQLIUQzSq8UpPvbprvXnEk/m1lU8dgSsAshWidHB+uadJn1IoQQomlJkC6EEM0oLb+icJx1untthePiMisC89hMCdKFaM2smXSjZNKFEEI0MQnShRCiGaXXUDiuoLScImN5lf3KTGYSs4ttP8dVyqoLIVofx0qF46RXuhBCiKYkQboAYPz48Tz44IMtPQwh2pVyk5nMQiOgFo5zMzjgrNcB1bPpSTnFlFcqFhcr093tkpJbgkmK7bVKixcvZtiwYbi7uxMQEMD06dOJjo6u85hly5ah0Wiq3JycnC7SiBvGmkk3K0qb+bcn13ohhGgbWjxIf+edd4iIiMDJyYkRI0awZ8+eWvctKyvj+eefJyoqCicnJwYMGMBvv/12EUfbOk2dOpXJkyfX+NzWrVvRaDQcPnz4Io9KCJFZaERRQKsBH1dHNBpNpeJxVYP0s+dlzmVNesNtOJHKyMXreXP9qZYeiqjB5s2bWbBgAbt27WLt2rWUlZVxxRVXUFhY979xDw8PkpOTbbe4uLiLNOKG0Wo1OOguXoV3udYLIUTH0aJB+sqVK1m0aBHPPPMM+/fvZ8CAAUyaNIm0tLQa93/qqaf44IMPeOuttzh27Bh33303M2bM4MCBAxd55K3L/PnzWbt2LYmJidWe+/TTTxk6dCj9+/dvgZEJ0bFZi8b5uRnQadUiU4Hu1jZsVau3W9ej9wxyV3/OKpI2bA30+5FU9f5oSguPRNTkt99+47bbbqNPnz4MGDCAZcuWER8fz759++o8TqPREBQUZLsFBgZepBE3nONF7JUu13ohhOg4WjRIf/3117nzzjuZN28evXv35v3338fFxYWlS5fWuP///vc/nnjiCaZMmUJkZCT33HMPU6ZM4bXXXmu+QSoKGAtb5tbANW5XX301/v7+LFu2rMr2goICVq1axfTp05k9ezahoaG4uLjQr18/vvzyy2b4ZQkhKrO2WrNmzwH8a8ukZ6iZ9NFRfjhoNRjLzSRLG7YGOZiQA0B0aj55JWUtOxhRr9zcXAB8fHzq3K+goIDOnTsTFhbGtGnTOHr0aJ37l5aWkpeXV+XWYI281juai9GUFWEsLpBrvRBCiCbj0FIvbDQa2bdvH48//rhtm1arZcKECezcubPGY0pLS6utSXN2dmbbtm21vk5paSmlpRUfhu26aAOUFcFLIfYd01SeSAJH13p3c3BwYO7cuSxbtownn3wSjUbN2K1atQqTycTNN9/MqlWrePTRR/Hw8ODnn3/mlltuISoqiuHDhzf3uxCiw7K1X3OrCNID3GvulW7NpEcFuBLu48KZjELOZhQS6uV8kUbbNhWUlnMyLR9QY52D8TmM7e7fwqMStTGbzTz44IOMGTOGvn371rpfjx49WLp0Kf379yc3N5dXX32V0aNHc/ToUTp16lTjMYsXL+a5555r3MAaea0Pb9yrVSXXeiGEEOdpsUx6RkYGJpOp2vS1wMBAUlJqnrI4adIkXn/9dU6dOoXZbGbt2rV8++23JCcn1/o6ixcvxtPT03YLCwtr0vfRWtx+++2cPn2azZs327Z9+umnXHfddXTu3Jl//OMfDBw4kMjISO677z4mT57M119/3YIjFqL9sxaHC3Cv+HKxtjZsZy1BeoSvKxF+rlW2idodTsypkojcF5fdcoMR9VqwYAFHjhzhq6++qnO/UaNGMXfuXAYOHMi4ceP49ttv8ff354MPPqj1mMcff5zc3FzbLSEhoamH3+LkWi+EEB1Di2XSG+O///0vd955Jz179kSj0RAVFcW8efNqnR4P6kV70aJFtp/z8vLsC9T1Luq33C1B79LgXXv27Mno0aNZunQp48ePJyYmhq1bt/L8889jMpl46aWX+Prrrzl37hxGo5HS0lJcXBp+fiGE/axT2v3d686km8wKCVlq+7XOvi509lX/35TicfWzTnXX6zSUmRT2x0uQ3lotXLiQn376iS1bttSaDa+NXq9n0KBBxMTE1LqPwWDAYDDU+nzdL9C4a31ecRlxWUU463V0DXBr/Gs3kFzrhRCiY2ixTLqfnx86nY7U1NQq21NTUwkKCqrxGH9/f9asWUNhYSFxcXGcOHECNzc3IiMja30dg8GAh4dHlZtdNBp1GlpL3CxT2Rpq/vz5fPPNN+Tn5/Ppp58SFRXFuHHj+M9//sN///tfHn30UTZu3MjBgweZNGkSRqPRvt+FEMIu1unuldekW7PqlTPpybnFGE1mHHVagj2d6WLJpMdmSK/0+hyMzwFgan91qvKB+Jw20w6ro1AUhYULF/Ldd9+xYcMGunTpYvc5TCYTf/31F8HBwc0wQhp9rXd0cUfRu1CqdZJrvRBCiCbTYkG6o6MjQ4YMYf369bZtZrOZ9evXM2rUqDqPdXJyIjQ0lPLycr755humTZvW3MNtE2644Qa0Wi0rVqzg888/5/bbb0ej0bB9+3amTZvGzTffzIABA4iMjOTkyZMtPVwh2j1b4bhKmfRAS8Beubp7nKX9WpiPMzqthghfV8v29ptJf3dTDAtW7KekzFTrPqXlJu778kCtrdUURbFl0m8YFoabwYGC0nKiU/KbY8iikRYsWMAXX3zBihUrcHd3JyUlhZSUFIqLi237zJ07t0qNmueff54//viDM2fOsH//fm6++Wbi4uK44447WuIt1MraK91kvni90uVaL4QQ7V+LVndftGgRH330EZ999hnHjx/nnnvuobCwkHnz5gHVL9q7d+/m22+/5cyZM2zdupXJkydjNpv5v//7v5Z6C62Km5sbN954I48//jjJycncdtttAHTr1o21a9eyY8cOjh8/zt///vdqMxiEEE0vvaCm6e5qJj2vpNwWoFZej175vr22YcsvKeP1P07y8+Fk/jhW+9+i9cfT+PFQEkvWnSSzoLTa88m5JaTll6LTahjQyYtB4V4A7JMp763Ke++9R25uLuPHjyc4ONh2W7lypW2f+Pj4KvVlsrOzufPOO+nVqxdTpkwhLy+PHTt20Lt375Z4C7XSaTW29ooXow0byLVeCCE6ghYN0m+88UZeffVVnn76aQYOHMjBgwf57bffbMXkzr9ol5SU8NRTT9G7d29mzJhBaGgo27Ztw8vLq4XeQeszf/58srOzmTRpEiEh6vTPp556isGDBzNp0iTGjx9PUFAQ06dPb9mBCtHOKYpSY+E4D2cHHB3UP73W6fDWTHpnS3Ae4uWEXqe2YUvKLaa92R6TQbnly4df/6q98OfPlufMihqwn8+aRe8Z5I6zo47B4d4A7Jfica2Koig13qzBJcCmTZuqtBZ74403iIuLo7S0lJSUFH7++WcGDRp08QffANZe6cbyixOkg1zrhRCivWvxwnELFy5k4cKFNT63adOmKj+PGzeOY8eOXYRRtV2jRo1COa/nqo+PD2vWrKnzuPN/10KIC5NXUk6p5UN75Uy6RqMh0MNAQlYxqXklhPm4EGspEBfhpxZ4ctBpCfNx4Ux6IXGZRXTybl+FnzZFp9seb4xOo8hYjotj1ctRsdHExhMVgflvR1O4YVjVop/WIH1gmBcAQzqrQbpUeBcXk16npbjMhPEiZdJBrvVCCNHetWgmXQgh2itrltzdyQEnva7Kc7bicbZMuhqkWzPpUDHlPbadVXhXFIWN0WrwrddpKCkzs+FE9Sz55pNpFBlNuBvU4H3bqQwKSsur7GMtGmcN0geGe6HRQHxWUbU+9EI0F+vMmIs13V0IIUT7J0G6EEI0A2uQWDmLbmVrw5ZXgtms2Ka7R/hWZMytQXp7a8N2PDmf1LxSnPU6bhkZAcCvf6VU2+9ny7ZZw8Po4ueK0WRmU3RFMF9uMvPXuVwA21p0Dyc9PQLdAdgfl9N8b0KISvQtMN1dCCFE+yZBuhBCNANb+7UagvRAj4pMemp+CaXlZhy0GkK9nG37dLFMfT+b2b7asFmz6KOjfJkxKBSADSfSKDZWVHkvKTOx4bha8GpKv2Cu6KPWKfntSEUwH52aT3GZmmmP9KvoTz3YMuVd+qWLi8XRwVo4rv0VeRRCCNEyJEgXQohmUBGkO1V7zppdT80r5WyGtf2aCw66ij/J1qnvZ2tow1ZQWs7y3XHVpn831tZT6fx5NqtJzlWfzZb16ON7BtA31IMwH2eKy0xVsuSbT6ZTaDQR4unEwDAvJvcJAtS17KXlajBvXY/eP8wTrbaiz/SQcFmXLi4uWyZdprsLIYRoIhKk1+D8YiyiceT3KDoy63rzOqe755dUWo9etThcFz81SI/PLKrWf/nFn4/z5HdHeP7Hoxc8zpi0fOYu3cOcj3aRkNW8WfvcojJbe7Tx3f3RaDRM6RsMVFRyh4qK71f2C0ajUdurBXoYKCgtZ0dMJlB9PbqVtXjcX4m5toBeiJo01TXKWt293GRuly0T6yPXeiGEaHoSpFei1+sBKCpqX9NLW4r192j9vQrRkdQ13T3AMt09Pb/UNp09olLROIAQL2e1DZvJTHKlNmx5JWWsOXAOgDUHkmyv01ifbj+LoqhTdd/acOqCzlWfrTHpmMwKXQPcCPNRv5SY0k8N0jecSKOkzERJmYl1lnZrU/qpGXStVsMkSzbdOuW9orK7d5XX6Ozrgq+rI0aTmSPn8pr1/Yi2qamv9TqtBq3m4vZKb03kWi+EEE2vxVuwtSY6nQ4vLy/S0tQPiC4uLmg0mnqOEudTFIWioiLS0tLw8vJCp9PVf5AQ7UxdheMCPazT3WvPpOu0GlsbtrMZFW3Yvt2XSHGZmiE2msx8sSuOhyZ2b9QYc4vK+Hb/OdvP3+w/x73juxLh51rHUY238YQ61f2yHv62bf07eRLq5cy5nGI2Raej02ooKC0nyMOJQZUC8El9gvh8Zxxrj6fyRFEZMekFQPVMukajYXBnb9YdS+bkiaMM8YoAnQEcHMHBCXSOIH/XO7TmuNbrzOWYTCYKiopQDB0jWJVrvRBCNB8J0s8TFKRma6wXb9F4Xl5ett+nEB1NXWvSrduyi8o4laYGm+dn0gG6+LpyJr2Q2MxCLunmh6IofLE7HlALr+04nckXu+K4Z3xUtTZvDbFybzzFZSZ6BrkT4uXMhhNp/Hf9Kd64caDd56qP2ayw+aQ1SA+wbddoNEzpF8RHW2P59UgyOkuwNLlvUJW15sO7+ODprCer0Mgn22NRFAj1cq74EsRUBsmHIG47T+au5VXDPjx3FsHO8wZi8IRLF8GoBaBrYDBlKodjayAnDrR69Titg3pz8YXe1zT21yJaSFNf6zMKSikpM1OWo8fV0LE+Wsm1Xgghml7HupI0gEajITg4mICAAMrKylp6OG2WXq+Xb9VFh2Zdkx7gUT2T7u2iR6/TUGZSiLEE6edn0gFbRjvO0oZtT2wWMWkFOOt1vDNnMFe/tY1zOcV8f/AcNw4Lt2t8JrPCZzviAJg3JoI+IZ5sOJHGmoPnuHd8FN0srcyaytGkPDIKSnF11DE0wqfKc1f2C+ajrbGsO5ZqC8yv6h9cZR+9TssVPX1JP/Qrkds+4H/6TMKUcnirHEryoCQHTEYAIgA0UIYDDjotGst2AEpzYd0zcHglXL0EwkfUPmhFgRM/w/rnIONkzfv49ZAgvQ1q6mv9mrUn+elwCjeP6My8S7o0wQjbBrnWCyFE85AgvRY6nU4uPEKIRiktN5FTpH7w93erHqRrNBoC3J04l6OuNddqsE1nr8zaN91a4d2aRZ8+KARvV0duGx3Bi78c55NtsdwwNMyuKbtrj6VyLqcYbxc90waG4qTXMalPIL8fTWXJulO8c9Ng+950Payt18Z09cPRoWo5lEFhXoR4OpGUW4wWhQB3Z1uVdgAyTsGBL/hX7HKcHNVsPDqg1HKzcvaG8NGUhY3kht80HC7vzKb/m0CYl5MawJtK1aD79ych7RgsvQIG3woTngWXql8cEL8b1j4NCbss5/aB7pNBMYO5TM3cm8vBI6Qpf03iImuqa72nuwvn8k2cyizFyan67BkhhBDCHhKkiw6t2GjC2VG+jBFNK6NAzdzqdRq8XGqeUu3vbrAF6aHeztUCV6jIpMdmFJKeX8pvR9Sq5zeN6AzAjcPDWLLuJCdTC9h6KoOx3f2rnaM2y3bEAjB7eLhtqvxDE7vzx7FUjh7ZT+bqL/Alp/qBrv7g2Um9eXRSg1StzhK8mlAUE+Vm0Lt4qlPLter7sgbpl/WsmOquTlE/jCZuO586/U5QyX48NUWUl+vRvuICehd1enmu+uWEE5CtuLPGNJrD5kjunTyIbuGhYPAAJ0/wDAOtFj2gHNqOKSGHJ9ccYXC4F2HeLoT5uOAXeg0l143Gd+cLBJ5eBfs/w3zgC7R6F3Xdus4AOgfIUV8TB2cYdS+MeUB9DSFqYP2SLTG7uJ49hRBCiPpJkC46rHc3xfDaHydZcccIRkT6tvRwRDuSlmcpGudmqDW7Xbnqe03r0StvT8gqZuWeOJxMhVweoqGvcgrOluLRaSgzh4axbMdZPtkW2+Ag/XhyHrvOZKHTarh5ZGfb9p6aRFb5fcygvI3ojjSurZIGsH4toaABJ0/MBi9ezFbQOJrputsF9ihqUJ+XDGXqLIEe1oMBB6UMSnLVG4BGB90mwsCb+Oe+AH46molOq2HxqElQy1r8S7r6cTAhhy0n09liWQtf1QyGa3rygn4p3bXnwJgPlWbFo9HCoJth/OOUOAdywwc78XJx5JNbh9r6YgthFerlDGD74k0IIYS4EBKkiw5r95ksTGaFffHZEqSLJmUtGufvUfu018BKz9W0Hh0gRJfLI/pVXKPZRtDWLBY6mSAL+Niyg0cn7htyHys0wWw+mc6p1Pz615KX5vPr+o301cQzPsqTkKw9kFIIB5fDiZ8YCqCBdaZB9Bp9DaGVp+ErJihIg7xzkJuo3vKTURQFk6KhXNFgRosWM06aMjQoUJKDriSH3ta4NvO88Th5QefRmMNHsfioD2kOQbx+bU905SVQVgRlxeDTBdzUDPwVZUn8dDSTfqGedRbLW/i3rvQK9uBsZiEJWUUkZBeRkFVMVqERN4MDHs4OmJ1Gc2vWAJT8ZB4cH8aswUFQXqrePELAKwyAg2cyOZyofmHw7sbTPDChW92/Y9HhdPJWg/SUvBLKTWYc5IscIYQQF0CCdNFhZRepabPMAmM9ewpRD0WBvCRIOw5pR+l6dA//05/FN98RPveo2M/FFzoNg07DCXJ1tm2ulkk/tx92vYfu6Hcs0FUtaqXoXdG4+KoZ6LxEfDc+yg7XQF4tnsqybcG8eN15a8lN5ZB0AM5shNMbUBL/ZJG5nEUGIAH4vPLOGuh9Da8VX81bx10ZFufN8okja5yKD5CYXcTj3/7F1lMZADg6aLl5RGeuHRzKLwfO8sveE+hKc/GiAFdNCVP6hzJreIQ6PV6jVdeQ+/UArRYt8OSY+n/VU/sHU2I0MbizV537Oel11YrP1eS9Taf5929GNmd4MCugV4377I/Ptj1+a8MpJvQOoE+ITH0XFfzdDDjqtBhNZlLySmqsMSGEEEI0lATposPKKrQG6aX17Ck6vNJ8OPk7HPse4naoU7U12opg01ikVg23iAQidUAJcOa8cx35BoC/aw0Md+xMquLNiOOucBY1g1ucrRY1szhl6MMb+Zez39yNaaP78vg1liC8rAT2fwZbX8evIIWX9R+TcXgV5UlBOFRqXUZ+csW0cdQZ5dmKGyadAV9PDzQOBrV3eFA/GH0/BPRkZmYRH8ds4c+z2Tz09UHenDUInbbqtP2ErCJmfbiLcznF6HUabhwWxoLLuhLsqX750Dd0APdO7MPqvQks3X6Wo/klPDn+Egi6sKrxGo2GG4aFXdA5KhsU7gXAwYScWvfZH6c+525wIL+0nH+sOsz3C8bU+uWF6Hi0Wg3BXk7EZRZxLrtYgnQhhBAXRIJ00WFlW4P0Qsmkd3jlpfDXajVAdjCoN51BrQZ+8neIWQflJXWfQ6MD364Q0Iu1mX78kqDnir7BXNnXks1VFLUYWeIeSPwTh+Jshmktbb2SzjuXVg99r4URd/PVASd+2aYWebtxVPeKffROMOLvMHguyt5Pyf7jFfzIhsxcqnHyhC7jOOw0hP/b78uJUh9ev2EA1w7uVONbCfd14YNbhjD/sz/5+XAyns56Xpze17a+vnKAHunnyrJ5wwmvYcq+m8GB28Z0Ye6oCEyK0irXcvcL9USrgeTcElJySwjyrLpEQVEUDiaomfRXbxjA49/+xfHkPN7eGMOiid1rOqXooDp5O6tBuqxLF0IIcYEkSBcdUkmZiUKjCZDp7h1e5mlYPQ+SD9W9n08U9JkO3a8Eg7u6PttS0RydI/hEqoEzsPKzP1lnTmNoVF/o37n6uRSFU8cP8N4XK3HXFPPUtEHoDc6WLwicIGQQuAcB0DslEYBLu/kR6e9W/Vx6ZzSj7iWt03U88fkqcgqLcdRpuOPSSMZ28weDOwXevXjupxOs2qmea2hn73qngo/t7s8bNw7kvi8PsGJ3PF7Oev5vcs9qAfpXd40koI6196BmGbU0vD3cxeRqcKBHkAfHk/M4mJDNZM+qv5eErGIyCozodRrGdffnX9P6smDFft7ZGMMVvQPpGyrT3oXKWjxuT2xWrV+ACSGEEA0hQbrokKw9rAEyC2W6e4d1eBX89CAYC9Q+2F0nqNnzcqOaOVdMEDYCek+HwD7QwD7k1sJxAe61BK8aDeHdB3AisJCoADf0wwfVeq5pA0NQgMt61F25vWdYIC89eBcPrjyoVjTfAHMKvZg+MJRH3t5BXGYRGg0sGN+VByZ0a1BW++r+IeQVl/PEd3/x7qbTmBX48VCSXQF6WzAo3IvjyXkciM9hct+qQbp1PXqfEE/bOvdf/grm57+S+ceqQ3y/cAwGB2njKOBvPQP4em8iX/2ZQJCnEw9OkJkWQgghGkeCdNEhZVWa4p5VaERRlFpbZYkmVJIHxVlq9llRyCks5s1NZxk/cgRjewTUf3xjpB2HlCNqZtojBNyDAQV+/T848IW6T+cxcO1H4Bla56kKSsvZeCKNib0D66wsnmJpwVa5zdr5DA46fnng0nqH76DTcv2QhmXlfFwd+fS2Yby14RT/XX+KFbvjWbFb7fcd4unEGzcOtLuTwZwR4eQUG3nlt2je33waoF0F6AADw7xYsTueA/E51Z47YAnSrWvXAZ6f1oddZzI5kZLPW+tj+MekHhdppKJFKAqk/KXWbajjOjG5bzBPTunFi78cZ8m6U2jQSCcAIYQQjSJBuuiQrJXdAcpMCnkl5Xg66+s4QlywlL/gkyvUtloWXsDTwMazIyh9/EcMjrUHtY1y/Ef4eq46Lb0ynSOYjIAGxj0K4/5PLQJXB7NZ4c7P9rLzTCb/vLo38y/pUuN+JWUmUvPUTLq1LdPFpNNqeHBCdwaFe/PAVwfIKSrj6v7BvDijX6P/jd8zLorcojI+2HKm3QXoAIMtAfjhcznV2mfttwTug8O9bdt83Qy8ML0v9yzfz/92xXHn2Ej5+9GepfwFH1wKboEQdTl0mwCRl4GLT7Vd7xwbiYLCS7+c4I11J9Fo4P7LJVAXQghhHwnSRYeUdV6xuMyCUvmQ3ZxM5fD9AjVA1zmCzoCi0ZBbasZNKeIydhO9/CF6zHu36V4zdgusvl0N0AN6q0F5XpI6BpMR3ILguo+hS/3ZbICl22PZeUZt8n0iOa/W/axFo1wcdfi4Ol74+2ikcd39Wb9oHHFZRQwK87qgmSIajYbHruzJlH7BdA1ww9XQvi4dkX5uuDs5kF9SzomUfNs682KjieOW/9aDO3tXOebKfsE8dmVPpg4Ikb8d7V3WadC7QEEqHFqh3jRaCB0Kk16EsOFVdr9rbBRmBV7+9QSvrz2JVgML/yaBuhBCiIZrX5+0hGigypl0UCu8R9a95FdciF3vqIXZnLxgwR5wD+T3I8nc/cV+Jmv38L7jEnrELce0qz+6kXdf+OslHYAvZ6vBeM+rYeZnoHNQp62W5EJBGniF2wq91edUaj6v/B5t+zk+q6jWfRMsz4V5u7T4EgpfNwO+bk0zO0Gj0TAgzKtJztXaaLUaBoZ5sfVUBgcTcmxB+l/ncik3KwS4GwjxrP5v5e5xURd7qKIl9JkBPaZA/E44tRZi1kP6cbVTw+Z/w83fVDvk7nFRKAr8+7cTvPrHSfp18mJcd7nICCGEaJjW1w9HiIugeiZdKrw3m8zTsPEl9fGkF8E9EICv/kwAIGjkDfxXcxMA2t8eh5N/XNjrZZyCL65Ti8FFXArXfaIG6KCuJ3X2Av/uDQ7QjeVmHvr6IMZyM+E+apuxhLqC9Gw1kx7mc/GnuovGG2T5AqLyunRr0bjB4d4t/oWLaGEOBogcr/4NW7ALblmjbk8+pH75V4N7xkcxY5Ba52JHTMbFGacQQoh2QYJ00SFlnx+kS4X35qEo8OMDaqX0yPEwUA3Gk3KK2XwyHYDbRkegGfMgK8vHo8GMsnqeWuitPsYiSDsB6dGQEaN+GZB8GP43A4oyIXggzFrR4GBcURSUGj5sv73hFEfO5eHloue9mwcDkJxXQmm5qcbzJFoC+E7e1fuGi9ZroGVd+gFLT3SA/XGWIL2zVwuMSLRqYSPUKe+F6ZCfUutuw7uo69aPJtW+REYIIYQ4n0x3Fx1SVqUWbCCZ9Gaz/zM4u1Vdz3n1Eltl5K/3JqAoMCrSlwg/V24d3YVxW+4kzJTGaOMxWHEjXP40eHZSbx4hgAaS9sOZzXBmkzrV1FTLfzffruoUVCePBg3zbEYh17y9DW9XR6b2D2HqgBB6BLlzID6bdzapFc1fmN6X3sEeuDjqKDKaOJddXGPf8oRsy3R3HwnS25KBYeqa8zPpheQWleHh7MCBhBwABoV713Gk6JAcXcCvhzrtPfkQeATXuFufEPVv0NGkXOkiIoQQosEkSBcdkjWTHurlzLmc4mrT3zuikjITW06mc1nPgAb10K5XXjL88bT6+LInwUethm4yK6zamwjArOFhAHi66LlxVBR3b36QXxyfp1NeInx3V6WTadTppuUlVV/D4AlaLShmFLMJs9lMrlskB4e/iTbRjKshCxdHHZF+bjg71l69/Y9jKeSVlJNXUs7bG2N4e2MMPQLdKSgtx2RWmDYwhKv7hwAQ7uPCiZR84rOKag7SsyzT3VugsrtoPB9XRzr7uhCXWcTBxBwi/VxJzy/FQauhn2WNuhBVBPdXg/SUw9Bjco27dA90R6fVkF1URkpeCcGe8ndBCCFE/SRIFx2SNSjvGuDGuZxiMgpkuvsLPx/ji13x3DqqM89N61v/AYWZaiGluB0Qt12dYu7kCc7e6n1OHJTmQshgGHmP7bBtMRmcyynG01nPpD5Btu3zL+nCp9vPMrPwUdb030lgWSLkJkLuOTCVqgG6szd0GQtdxmGOGMehIh92nMlie0wGe+OyMZaboQD4LgWomIIa5uPMxofHV2mtVdmhxFwAJvcJotyssPlkGtGp+QAEeTjx/DUVv4/KQXpNJJPedg0K81KD9PgccizFJfuEeOCkr7s9n+iggvrD4ZVqJr0WTnodXf3diE7N5+i5PAnShRBCNIgE6aJDslZ37xbgxuaT6R1+untaXglf/6lmt5fvjmfemC5EuJVDwp+QdkytiF6aByV56n1WrJpBOl9uQtWftQ5wzVtVepCv/DMegBmDQqsEPwHuTtwwtBNf7DLzj+Lb+N/8EeoTZjMUZaiv7ROpZs6Bf3x9kG/3V1RcBzWg7hboRmmZmUJjOUVGE/FZRSRkFVdprXW+vyxB+i2jOjOmqx+5RWX8fiyF3WeyuG10BJ4uFS22rMXj4jOrB+n5JWXkWJZSSJDe9gwK92bNwSQOJGTb/kbIVHdRq+AB6n3K4Tp36xPiQXRqPseS85jQO/AiDEwIIURbJ0G66HAURbFl0rsFqtOVO1zhuNMbYP2/1DXinp2Iy3LhZhzI1znTTxOLw4dPgjEWqLlqsY1/L+g8CjqPAe8IKMmB4hzLfbbaRzioIgudUVDK2mOpANw4LKza6f4+Noov9ySw9VQGfyXm0q+TZTq7W4B6s8gqNPLDwSQArugdyCXd/BjT1Y9IP9dqaz5v+WQ3W09lsD8+u8YgPafIaMuK9w1Rn/d00XPD0DBuGFp9jOG+liC9hky6daq7t4set3bWS7wjGGip8H4wIcf2N2KQpaCcENUE9VPvc+KhKAtcfGrcrXeIB98eOMfRpNyLODghhBBtmXyKFB1OcZmJ0nIzAF0D3IHqLdnatX3L4KdFoFiqk5/bxzBgmL7SPtZfh3eEOl3d1Q8MHmohNoMHuAVC2HB1ux2+3Z9ImUlhQJgXvYKrF3UL83HhmgEhfHfgHB9tPcObswfVeJ6fDidRblboG+rBh3OH1vmag8O91SA9Lpu5oyKqPX/YkkWP8HWpkjGvjTVDXmOQLlPd27RewR44OmjJKSojp0j9dzFYMumiNs5e4NVZXdqT8hdEjqtxt96W4nHHkqXCuxBCiIaRIF10ONaA3OCgtRX3yio0YjIr6LTtuPKu2Qzrn4PtS9Sf+98Iva5h+/6DHDl+jB5OuYzrpGVLXiBfpYSg7TyCd/5+VZO9vKIott7os2rIolvdPqYL3x04x69Hksko6I2fm6HaPt8dOAfAjEGd6n3dwZ3VIGt/pf7Xlf11Tg3G+nfyqvdcQJVe6edXa7b2Tw+T9mttkqODlr4hHrZ/K35uBjpJAUBRl+D+liD9cO1BuuULyYSsYnKLy/B0rv/LQCGEEB2b9EkXHU52obpm2MfVEW9XRwDMCrZCUa1aaQFknYGEPXD8J9j7KRxdo67XrktZMXxze0WAPv5xmPEBpd2u5KGzI1lcfhNpk99Hc9uPRN78Jus1o/g5FraeSm+yoW+PyeRMeiEujjqmDgipdb9+nTwZ0MmTMpPC13sTqj0fm1HIgfgctBqYOqDmtkeVWacwx2cV1Vgg8HBiDgD9OzWsgneolzMaDRQaTdVmYCRmq9PdO/lIYNdWVV6DPjjcS1pmibpZ16Un174u3cvFkVAv9W/CccmmCyGEaADJpIsOJ8sSjHu7OKLXafFy0ZNTVEZWoRHfGrK2rULWGfj+PojbVvPzWr1a9bznFOgxRZ2Snhmj3jJOQvSvaqZHq4dpb8OAWQB8fyCJtPxSgjycmD4wFFCnat8yqjOfbIvl5V9PMCbKD209Mwy2x2SQWWjkmlqCb0VRWLLuJAA3DA2rd732TSM7c2j1YVbsjufvY6OqzHBYY8miX9rNnwB3pzrPA+DprKdbgBun0grYH5fNFZUqykPFdPeGttly0usI8nAiObeE+KyiKv9mJJPe9lVegy5F40S9gqxBeu0V3kGd8n4up5ijSXmMjPS9CAMTQgjRlkkmXXQ41h7pPpYsuvU+ozVWeFcU2PcZvHdJRYDu4Axe4WpRtu6TwbcbmMvg9Hr4+WF4vRcsDoUPx8E382Hzv9UA3ckTbvnOFqCbzQrvbzkNqO3PHB0q/hwsuKwr7gYHjibl8ePhpDqHWG4yc9fne7n/ywO2onDn2x6Tyd64bAwOWu4ZH1Xv257aPwQPJwcSs4vZcrIim68oCmsOqkH6tYND6z2PlXVd8flT3tPyS0jOLUGjodbK7zWpbV26rElv+6wzL0DNpAtRp+D+6n3mKTDW3JYR1ArvgBSPE0II0SCSSRcdjnWKsnWqu5+rgTPphS1b4d1UDrrz/ncsSIcf74foX9SfO18C099Ri7mdL/0kRP8MJ36BxD8BBVz8wK87+HVV73tPU4N7i7XHUzmTXoiHkwOzR4RXOZ2PqyN/HxfJq3+c5NU/ormyb3CVIL6y+KwiCo1qEbpnfzjKmK6+uDhWvBdFUXjDkkWfMyKcQI/6s9/OjjquHxLG0u2xfLErjst6qpXd98fnEJdZhIujjol2tDIa3NmLlXsT2B+fXWW7tfVaV383XO2oxt7Zx4U9sVlV2rApimKr7h4m65jbrFAvZ8Z29ycjv5QBlQJ2IWrkHqQW0ixIhdSjEDasxt36WDpHHEuS6e5CCCHqJ0G66HCs/Y99LJW8fd3UYL3FeqXveh9+ewz0zuDqX3E7txcK00HnCH/7J4xaUKXfeBX+3dXbJQ+prYCg1nZAoAaU729Ws+i3jOpc4/Tz2y/pwmc740jIKua3oym1TmWPSSuwPT6XU8x/15/i8St72bZti8lgnzWLPq7+LLrVTSPDWbo9lg3RaSRmF9HJ24XvDqi93Cf3DaryRUB9rJn0w4k5lJnM6HVay8/2FY2zCq8hk55VaKS4zIRGA6ESpLdZGo2Gz28f3tLDEG1JUH+IWQsph2oN0q0V3mPSCigtN2FwqOVvuRBCCEErmO7+zjvvEBERgZOTEyNGjGDPnj117r9kyRJ69OiBs7MzYWFhPPTQQ5SUlFyk0Yr24PxMui1Ib4k2bAVpsOFfgAJlRWqV4HN74eSvaoAe0Bvu3Ahj7q89QD+fi0+dATrAb0dSOBCfg6ODlttGd6n5NI4OTOmrrt8+cq72KZqnLEG6tQr2J1tjOZGiZosUReGNtWoW/aYRnQloQBbdKsrfjdFRvigKfLknHmO5mZ8OJwMwY1DDp7pbz+Xh5EBJmZnolHzb9orK7g2f6g4190pPsBSNC3R3kg/gQnQk1invdaxLD/F0wstFT7lZ4VRqQa37CSGEENDCQfrKlStZtGgRzzzzDPv372fAgAFMmjSJtLS0GvdfsWIFjz32GM888wzHjx/nk08+YeXKlTzxxBMXeeSiLcuuVDgOwMdVLfyVWUPl72a36WUwFqi9yO8/APPXwqwVcPUSmPauGqAH9W3Sl/xiVxwLVuwHYM7wcPzday+W1yNIzf6cqBTYnu+0JUifPTycK3oHUm5WeOq7I5jNCltOZbA/PgcnvZa7x0faPdabR3YGYOWfCaw9lkpOURkB7gZGR9nXn12r1TDQti5dnfKuKIqtsns/O4P0sEpt2KxsReOksrsQHUuQNUivvcK7RqOxtWKTdelCCCHq06JB+uuvv86dd97JvHnz6N27N++//z4uLi4sXbq0xv137NjBmDFjmDNnDhEREVxxxRXMnj273uy7EJVVW5PeUtPd00/CvmXq4yv+BT6REDYcel4FQ+fBoJtA3/DMc33MZoWXfz3BU2uOYFZg5pBOPHlVrzqP6RHkDkB0Su3rKK2Z9K4BbjxzTR9cHHXsjctm1b4EW0X3m0d0blAl9vNN7B1IgLuBjAIjz/xwBIBpA0Ma1c/eWgRsf5wapCfnlpBRYMRBW/HhuaGs092T80ooLVfX49uKxklldyE6FmsbtrRjYCqrdTdr8ThZly6EEKI+LRakG41G9u3bx4QJEyoGo9UyYcIEdu7cWeMxo0ePZt++fbag/MyZM/zyyy9MmTKl1tcpLS0lLy+vyk10bLY+6ZZMuq8lk35+z+tmt/45UEzQ/UqIuKRJTllsNHHjBzuZvGQL//rpGBtOpFJQWk5JmYkHVh60rUNfNLE7r1zf37Y2uzbdA90ASM0rrbGPvNmscDq9IkgP9XLmoQndAXj6+6McsGTR/27HWvTK9Dots4aFARXV92cM6tSocw06r8K7dT1690B3nPT2TU/3dXXExVGHosA5yzR3a9G4TlLZXYiOxTsCDJ5gMkJ6dK279bZVeJfPIUIIIerWYoXjMjIyMJlMBAZWrdAcGBjIiRMnajxmzpw5ZGRkcMkll6AoCuXl5dx99911TndfvHgxzz33XJOOXbRttj7prlULx2XUUt3996MpJOUUM29MzWu3GyVuB5z4CTRamNh0/z43RaexO1YtHHciJZ9PtsXioNXg52YgJa8EB62Gl6/rz/VDGhboujvpCfVy5lxOMSdS8qv1903KLabIaEKv09DZEpzeNiaCb/Yn2qbI3zKyc51T6usza3g4b2+MwaxAzyB32wdde1lba8VnFZFRUGqb6m7venRQp66G+7hwIiWf+KwiIv3dSLRl0mW6uxAdikYDQf3UNpnJh2pdomSt8H48OQ+zWUHbiBlBQgghOoYWLxxnj02bNvHSSy/x7rvvsn//fr799lt+/vln/vWvf9V6zOOPP05ubq7tlpCQcBFHLFobRVGq9Un3da19urvJrPDQyoM89+OxOoun2TkI+OOf6uPBc8G/R9OcF7WtGsDlPQOYPTyMMB9nys0KKXkluBsc+Oz24Q0O0K16Wqa8n0ytvi7dWtm9i58rDpasvF6n5cUZ/dBowMVR1+gsulWIlzOT+qgF7Owde2Weznq6BagzA/bHZVcqGufVqPOdvy69Yk26ZNKF6HCsU95Tal+XHunnisFBS6HRRFylehZCCCHE+Vosk+7n54dOpyM1NbXK9tTUVIKCgmo85p///Ce33HILd9xxBwD9+vWjsLCQu+66iyeffBKttvp3DgaDAYOh8Vk80b7kl5ZTblaAisJxvm7qv4/c4rIq7bkAzmYWUmTpAb4nNou+ofZnXas5tkat4K53hfFNV/TQZFbYeEItunjX2EhGWLLe8ZlF7I3LYnC4NxF+rnaft0eQO+tPpNVYPM4apHcLcK+yfUhnb1b9fRQujg74uV34/38vX9efqQNCbMF6Yw0O9+ZUWgH74rMrtV9r3H/Tym3YTGaFczmWHukSpAvR8QTXXzzOQaelZ5A7hxJzOZqUS5dG/D0WQgjRMbRYJt3R0ZEhQ4awfv162zaz2cz69esZNWpUjccUFRVVC8R1OnUtqaIozTdY0W5Ys+gujjrbOmQvZz3WWYfZ561LP1kpMN0bl2X/CxbnQOZpyDgFaScg9Siss0xvH30fuAfWeXhlx5PzWH88tdbn98dnk11UhqezniGdvW3bw31duHZwp0YF6FC5eFztQXqUJUNd2dAIn0ZPTT+fp7OeKf2CG1UwrrLBnb0A+OlQMrnFZTjqtHQPdK/7oFpYg/S4zCJS80ooMynodRqC7GgzJ4RoJ6wV3lP+ArO51t16W6a8S/E4IYQQdWmxTDrAokWLuPXWWxk6dCjDhw9nyZIlFBYWMm/ePADmzp1LaGgoixcvBmDq1Km8/vrrDBo0iBEjRhATE8M///lPpk6dagvWhaiLrbK7JYsOansuH1dHMgqMZBQYq/Tyrpw93ns2G0VR0GgaGCge/wm+uQPKi6s/5xqgBukN9O3+RB795jBlJoWVd420ZckrW3dMDeD/1jPANvW8KViD9JMp+dXe/ylbJr16kN4aDbYUj7NmvXuFeODo0LjfVeVe6dap7iFezhf8RYIQog3y6w4OTmDMh+xYtVtHbiIk7IaSXHVpk04vxeOEEEI0SIsG6TfeeCPp6ek8/fTTpKSkMHDgQH777TdbMbn4+PgqmfOnnnoKjUbDU089xblz5/D392fq1Km8+OKLLfUWRBtj7ZFuXY9u5euqtvnKPK94XOXscVp+KQlZxbbgrE6px+Dbu9QAXe8KOr1aJE6rUz/ITXoRDPUHtoqi8Mbak7y5Ica27bOdZ2sM0q3r0Sf0anh2viEi/dxw0GrILy3nXE4xnSwtxhRFsWXSu7aRID3K3w0PJwfySsoB6H8ByxfCK61JT7BUeJf2a0J0UDoHCOwD5/apf/vzkiA/qeJ5BycYdFNFG7ZkCdKFEELUrkWDdICFCxeycOHCGp/btGlTlZ8dHBx45plneOaZZy7CyER7lGVpv+Z9fpDu5gip1duwWYulOTpoMZab+fNsVv1BelEWfDUbygqhy1i4+Tv1A5ydSspM/N/qw/xwSP2gd+2gUL49cI7fj6aSnFtMsGdFFfEz6QWcSS9Er9Mwtruf3a9VF0cHLVH+bkSn5nMyNd8WpKcXlJJbXIZWQ5tZW6nVahgY7s2Wk+kA9GvkenSAUC9nNBooNJo4lJADQJiPVHYXosMKHqgG6ef2qj9rdODkAcXZloJyN9EzyB2NBtLzS0nLLyHAXZbHCCGEqK5NVXcX4kLZKru76KtstxaPy6hU4b2kzMTZzEIAruoXDMDeuOy6X8BUDt/Mh+yz4BUO1y9rVICeWVDKTR/v5odDSThoNbxyXX9ev3Egw7v4YDIrrNgdX2X/9cfVgnEjI31xd9LXdMoLYp3yXnn6vzWLHu7jYnef8ZY0ONzL9nhAIyu7Azjpdbb159tPZwDYvsAQQlRYvHgxw4YNw93dnYCAAKZPn050dO39xK1WrVpFz549cXJyol+/fvzyyy8XYbQXYMwDMGQeXP403PYzPJ4AEy3dZ9LV1rIujg5EWr7UlHXpQgghaiNBuuhQKnqknz/d3dqGrWK6e0xaAWYFvF30XNlXrSq+96yleFzsVjjxs1oYrrL1z8HpDeDgDLNWgGv1aen1OZdTzMz3d7IvLhsPJwc+v304NwwLA+C20REAfLknntJyk+2Y5prqblVT8biKqe6NK7zWUqzr0p31OqL8L2wGgLWS+5n0wio/CyEqbN68mQULFrBr1y7Wrl1LWVkZV1xxBYWFhbUes2PHDmbPns38+fM5cOAA06dPZ/r06Rw5cuQijtxO3p1h6hK49GGIuAQcXStabKaftO3WM0id8l5TW0shhBACWsF0dyEupopMes1BeuXp7tascY8gd1u19NNpeZT88hROe95Sd9LoIHQIRF0GehfY8aa6ffo7ENTP7vGdSS/g5o93k5RbQqiXM5/dPrzKeu+JvQMJ8nAiJa+EX/9KYfqgULILjbYvDy7vFWD3azZEj8C6gvS2sR7danSULzePDKd3sOcFF9gL93FhT2xF1f8wb5nuLsT5fvvttyo/L1u2jICAAPbt28fYsWNrPOa///0vkydP5pFHHgHgX//6F2vXruXtt9/m/fffb/YxNxm/7up9fhKU5IGTB90D3fn5r2SiUwpadmxCCCFaLcmkiw7FVt292pr06tPdo1PUqYg9At3xdTMwwM/MMv2/KwJ0r3BQTJC4Bzb/G9ZZaiVc8hD0vc7usR1PzuOGD3aSlFtCpL8rq+4eVS0A1uu03DQiHFALyAFsOpmGWYGeQe7NNt3amkk/nV5AmUltL3QqtW1Vdrdy0Gl5YXo/5lh+jxci/LzMuWTShahfbm4uAD4+PrXus3PnTiZMmFBl26RJk9i5c2etx5SWlpKXl1fl1uKcvcBNnYlFhppNt3XMkEy6EEKIWkiQLjqU2qq7W3+uXN092hKE9gjygNRjfGJ8lLG6vyjTGuD6pfDgX/DgEbjmLegzQ22r1uda+Ns/7R7XvrhsbvxgJxkFRvqEeLDq76MI8ao5KztreDh6nYYD8TkcTsxh3TF1PfrE3s0z1R2gk7czbgYHykyKbWp3THrbzKQ3pcpBurNeZ5uRIYSomdls5sEHH2TMmDH07du31v1SUlJsnV6sAgMDSUlJqfWYxYsX4+npabuFhYU12bgviL8lm56ursO3Bumn0vIxmZWWGpUQQohWTKa7iw6lpj7pAH5u6s/js7+F9x8FgwfzE4xMdXBm3NkusG41fmWFJJj9+a/vs7xqzZR7han9bwfPbdR4FEXhh0NJPP7tXxQZTQzt7M0ntw3D07n24m/+7gau6hfMmoNJfLQ1ls2WSuXNtR4dQKPR0D3Qjf3xOUSn5hPk4UR6vvqFRlQHDtIrZ87DfJyr9JAXQlS3YMECjhw5wrZt25r83I8//jiLFi2y/ZyXl9c6AnW/HhC7BTLUID3cxwWDg5aSMjPxWUVtpjuGEEKIi0eCdNGhZBepLdiq9Ul3M9BNk8gC41JIUadzjwP1/5DjWwAo7nQJU2NuoSjVkxfLTRgcLqyi+aGEHJ7/6Rj7LBXjx3b354Obh+DsWP95546OYM3BJH60tGcLcDfQ7wJ6fjdEjyAPNUhPySPUS61qHuLphJuh4/4Z6VypHZ/0SBeibgsXLuSnn35iy5YtdOrUqc59g4KCSE1NrbItNTWVoKCgWo8xGAwYDIYmGWuTOq94nE6roVugG0fO5RGdki9BuhBCiGpkurvoMExmhRxbdffzWrC56nnOYRkOGjOmyL9xauxbPFE2n/ccbobR98OUV3GatwYHN1+M5WaOnMtt9DhS80p4+OtDTHtnO/visnFx1PGPK7rz8dyhDQrQAQaFedG/Uo/vy3sFoNU2bxa3R6CaMY9OybetR+8a2LYquzc1X1dHXCz/zWQ9uhA1UxSFhQsX8t1337Fhwwa6dOlS7zGjRo1i/fr1VbatXbuWUaNGNdcwm48tSD9h29QjUCq8CyGEqF3HTYGJDievuAzr8r/zp7u7n/6Z0bpjlCh6csYtZmeSEytMvlzeKQCuGAaABhjS2Zvfj6by59lshnSuvehRTQpLy/l4aywfbDlNkVFtn3btoFD+b3JPgjyd7DqXRqNh7qgI/rHqENC8U92teljaBp1IySfCV838dPXvuFPdQf3vEO7jwomUfDpJZXcharRgwQJWrFjB999/j7u7u21duaenJ87O6v83c+fOJTQ0lMWLFwPwwAMPMG7cOF577TWuuuoqvvrqK/bu3cuHH37YYu+j0fwsQXpOHJSVgN6JHkGWLz0lSBdCCFEDyaSLDsPaI93dyQF95dZbxkI0fzwJwHvl15CuC7a1X+seVDVTPCxCDcxt/dIboNxkZsXueMa/uok31p2kyGhiULgX3907mtdvHGh3gG51df9gIv1cCfNxZkxXv0adwx49Lb+LxOxiDibkANAtsGMH6QAjI33RamB4F/u+tBGio3jvvffIzc1l/PjxBAcH224rV6607RMfH09ycrLt59GjR7NixQo+/PBDBgwYwOrVq1mzZk2dxeZaLbcAcPIExQyZMQB0t8xCOpkiQboQQojqJJMuOgxbj/TzK3BvfQ3yzpGiDeR901QGFpbaPjj1PC9IH2oJ0vfFZWM2K/VOMd9wIpWXfjlh6yke5uPM/03qyVX9gi94erqTXscvD1xqe9zcvF0dCXA3kJZfyr54dR19R67sbvXM1N48NKE7ni61F/sToiNTlPormG/atKnatpkzZzJz5sxmGNFFptGAf09I2K0Wjwvqa6vwHptRSGkT1DgRQgjRvkiQLjqMGiu7Z56GHWrf86+876a0yJGM/FLbFMQe5wXpfUI8cNJryS4q40xGAV0Dal+TfSA+m/mf7UVRwMtFz/1/68ZNI8Ob9MPYxQjOK+sR5E5afinWz9wdfbo7qFPeJUAXQtTJr7sapFvasAV5OOHu5EB+STln0gvpFezRwgMUQgjRmsh0d9Fh1Ngj/bfHwGSEqMuJ878MgKNJeeSXlOOg1RDpVzUI1eu0DAzzAmDv2ew6X29PbBaKok6D3vzIZdx+SZc2ny2pPLPAz80Rb+kLLoQQ9bMVj1ODdI1GY/t7KsXjhBBCnE+CdNFhlKefZoz2LyYYN8K2JfDDfXDqD9Dq4cp/4+umtu7ZcToDgEh/Vxwdqv8vMtRSMO7PeoL0s5mFgLpmua6+521J90rV3GWquxBCNJC1eFzGSdsm69/TaFmXLoQQ4jwy3V20f+VG+OlBbjq4nJscgSTLzWrUveDXDV+30wCctLQXs1YzP9/QCG8A9sXVXTzubEYRABG+7ac1V89KvxMJ0oUQooGsmfSMU2AqB52DbTmVZNKFEEKcT4J00b4VZ8PKW+DsVsxoiTEH4+wTSlh4F3ALBO/OMOgWQO15Xdn5ReOsBoWrQfrZzCJyi8tqzZLHWTLpEX6uTfVuWly3QDe0GjAr0K2O9fhCCCEq8QwDvQuUFUH2WfDrWpFJlyBdCCHEeSRIF+1X9llYPlOdXujoxn+9nuC/8RG8PLofs4aHV9vd161qkF55andlns56gjycSMkr4XR6AYMtQXtlJWUmknJLAGw9xdsDJ72ObgHuRKfmS6EjIYRoKK0WfLtCymG1wrtfV3pYrjEJWcUUlJbjZpCPZEIIIVSyJl20T4l74aPL1QDdPQRu/40tDASotdiZdU26VW2ZdICoADXwPm1prXa++Cx1qru7kwPe7azy92s3DGDxtf0YFlH9ywkhhBC18O+p3luKx1nbWgKckmy6EEKISuRrW9H2GYsg9Qik/FVxn3QQzGUQ1B/mrASPELILNwI19Em3qDzd3dVRR6iXc60v2dXfje0xmcSk1xykx2aoU927+Lmi0VxYP/TWpm+oJ31DPVt6GEII0bb4d1fvLUE6VLS1PJmab1tKJYQQQkiQLtq2jFOw7CooSK3+XPcr4bqPwaAWOKuxT3ollae7dwt0R6utPbiOshRNqy2Tbl2P3rkdTXUXQghxAWwV3iuC9O6B7mw9lUF0Ss3XEiGEEB2TBOmibVv7jBqgO/tA6GAI7AtB/dSbX3ewZLHLTGbySsqB2jPpLo4OOOt1FJeZ6pzqDmomHeB0emGNz8daKrt3aUeV3YUQQlwAW6/0k6AooNFIhXchhBA1kiBdtF3xuyD6Z9Do4PbfK6YS1iCnqAxQY/a6epb7ujmSmF1s++BUG2v7sbjMQkrLTRgcdFWel0y6EEKIKnwiQesAZYWQmwheYbbicSekV7oQQohKpHCcaJsUBdY+rT4efEudATpAdpE61d3LWY+ujmns3SzB99DOPnWez9/dgLvBAbNS0Q+9srhMS4/0dtR+TQghxAXQ6cEnSn1smfLeLVC95mQUlJJZUNpSIxNCCNHKSJAu2qboXyBhNzg4w7jH6t3dth69lqnuVktmDeKn+y6hX6e6C6NpNJqKdennFY9T268VAxAh092FEEJY2YrHnQTUZVbhPup14mRqxbUko6CU2z7dw+PfHr7oQxRCCNHyJEgXbY+pHNY9pz4edS94BNd7SLYlSPeppWiclaezvsGVy6Ms69Jjzisel5BVhKKo7ddqW/8uhBCiA6qleBxUrEtPzC5i5vs72RSdzpd7EsgtLrvowxRCCNGyJEgXbc+hFeoHHGcfGPNAgw7JKmpYJt0eXWvJpFvbr0X4tr/2a0IIIS7Aeb3SAVuh0ujUfGLS8pn5/k7bdQSqfxEshBCi/ZPCcaJtMRbBxpfUx2MfAaeGZb0bmkm3R5S/ut78/A9Q1vXonWWquxBCiMpq6JXe20dhjPYvkqNTmPlXMtlFZXQNcMPgoOVoUh6n0wsY0ll6qAshREciQbpoGxQFTEbY9Q7kJ4NnOAyb3+DDt5zMACC8CQPnypl0s1mx9VWPtVR27yJF44QQQlTm2w3QQHEWfHMHJB9iSsZJpjhCabGeS0qX0L9TBMvmDefN9afUIF0y6UII0eFIkC4uvvSTcOIniLoMggfaepnblBvh2Br48xNIOw7lJWA6r+rt354CB0ODXu5oUi57zmbhoNVw/ZBOTfIWAMJ9XNDrNJSUmUnKLaaTt/oFgLRfE0IIUSNHF/AKh5w4+GuVbXO5osWgKWNOcDJ33jkLN4NDrbO1hBBCtH8SpIuLy1QOX82BzFOw/jm1HU2/66Hv9WBwh32fwt5PoTCt9nN0GQf9Zjb4JT/bcRaAK/sFE+jhdIFvoIKDTkuEryun0gqISSuwBenWlmxd/GS6uxBCiPNMfA7+Wg0BvaHTUAgdwtmvHqNrwiru65mHg0H9aFZbBxEhhBDtnwTp4uL662s1QNe7qFPYs07D5n+rN40WFLO6n3swDJ0PvaaCoys4GCjT6Hn6p1MYNQbuySika4B7vS+XXWjk+4NJANw2unOTv52uAW62IH18j4Aq7dckky6EEKKaPjPUWyVdB42DhFU4JB+o2GYJ0uOziigpM+Gk113UYQohhGg5EqSLi6fcCJteVh+Pe1RdU37iFziyGk5vAHM5hI2EEXdBr2tAp69y+Dd74vnyQDoA3x5IZNqAEB6Y0L3Otd9f/ZlAabmZvqEeDA5v+sI71jZsp9PVKe6J2Zb2awYHfKX9mhBCiIYIHaLeJx0Aswm0OvzdDLg7OZBfUs7ZzEJ6Bnm07BiFEEJcNHYH6REREdx+++3cdttthIeHN8eYRHt18At1HZ5rAAy/U82QD7hRvRVlQWk+eNec7TaWm3lrQwwA3QPdOJlawJqDSfx4OJkZg0J5aGJ3Qr2cqxxTbjLzxa44AG4dFdEs7dBsxeMsawZjLVPdO/u5SPs1IYQQDePfA/SuYCyAjFMQ0BONRkPXADcOxOdwOk2CdCGE6Ejs7pP+4IMP8u233xIZGcnEiRP56quvKC0trf9A0bGVlcDm/6iPL31YDdArc/GpNUAH+GZ/IudyivF3N/DDwkv4ceElXN4zAJNZYfW+RKa9vb1acZ11x9M4l1OMj6sjUweENPU7Aipn0tXXlqJxQggh7KbVQcgg9fG5fbbNXS3XGCkeJ4QQHUujgvSDBw+yZ88eevXqxX333UdwcDALFy5k//79zTFG0R7s+xTyk8AjFIbcZtehxnIzb1uy6PeMi8JJr6NfJ08+uW0YaxaMoWeQOxkFpcz+aFeVDzLLdsQCMHt4WLOt5YsKUIPxzEIj2YVGYjMs7dckSBdCCGGP0MHqfaUg3Vo8LkaKxwkhRIdid5BuNXjwYN58802SkpJ45pln+Pjjjxk2bBgDBw5k6dKlKIrSlOMUbZmxELa+pj4e+wjo7auwvnqfmkUPcDcwZ0TVJRYDw7xYcedIega5k56vBuqn0ws4kZLHrjNZ6LQabh7Z9AXjrFwcHWzT7GPSC4jLtEx3b8J+7EIIIToA67r0GjLp0itdCCE6lkYH6WVlZXz99ddcc801PPzwwwwdOpSPP/6Y6667jieeeIKbbrqpwed65513iIiIwMnJiREjRrBnz55a9x0/fjwajaba7aqrrmrsWxHNbc+HUJgO3hEw6Ga7DjWWm3lnoyWLPj6qxoy4j6tj1UD9w1288ls0AJP6BBLs6VztmKYUaellezqtgLOW6e51FbMTQgghqrEG6alH1CViVNQ9OZNRgNksyQ8hhOgo7C4ct3//fj799FO+/PJLtFotc+fO5Y033qBnz562fWbMmMGwYcMadL6VK1eyaNEi3n//fUaMGMGSJUuYNGkS0dHRBAQEVNv/22+/xWg02n7OzMxkwIABzJzZ8L7Z4sIpisLaY6n0DfUkxFqwLfUo/PGU2u/cu4salHuFwfb/qs+Pe6xaxXar0+kFHEvKY0KvQJwdKwLxVfsSbFn02cNrL1To4+rI8jtGcNPHuzmRks+GE2qf9VtHRTTF261T1wA3tp7K4FhyHkk50n5NCCFEI3h2UgurFqZByl8QNoxO3s446rSUlJk5l1NMmI/M0hJCiI7A7iB92LBhTJw4kffee4/p06ej11cPurp06cKsWbMadL7XX3+dO++8k3nz5gHw/vvv8/PPP7N06VIee+yxavv7+PhU+fmrr77CxcVFgvSLbHtMJnf9bx/Du/jw9d9HQXYc/O9aKEip+QC/7tD/hhqfMpabuemj3aTkleDr6sjtl3ThllGdMThoeceyFv3eWrLolfm6GVh+xwjmfLSb6NR8ega5M7yLT53HNAVr8biN0WmYFXAzOODnJu3XhBBC2EGjUbPpJ39Vp7yHDcNBp6WLnyvRqfnEpBdIkC6EEB2E3UH6mTNn6Ny57jW+rq6ufPrpp/Wey2g0sm/fPh5//HHbNq1Wy4QJE9i5c2eDxvPJJ58wa9YsXF1rzlyWlpZWqT6fl5fXoPOKuh1KzAFgX1w2+dmpuH9xnRqgB/SGQbdAdixkxUL2WSjJgckvq9Vra/DDoSRS8tSpfZmFRv7zezTvbz7N0M7eJOWWEOhhYFYdWfTKfN0MrLhzBB9vi+Xq/sEXpQ2adTpiQpY1iy7t14QQQjRC6OCKIN2ia4Ab0an5nE4r4LIe1WcYCiGEaH/sDtLT0tJISUlhxIgRVbbv3r0bnU7H0KFDG3yujIwMTCYTgYGBVbYHBgZy4sSJeo/fs2cPR44c4ZNPPql1n8WLF/Pcc881eEwdzq734Mwm6DYRek0DN/8GHXYqNR8AvbkE8xc3QOYp8OgEN60Gz9AGv7yiKHy89QwAD0/sTqi3M+9uOk1MWgEbo9MBuHd8V7uqs/u6GXh0cs/6d2wi1ky6VYRMdRdCCNEYNVV4t9Y9kQrvQgjRYdhdOG7BggUkJCRU237u3DkWLFjQJINqqE8++YR+/foxfPjwWvd5/PHHyc3Ntd1qGnuHVVYMa5+Gk7/Bzw/Daz3g82mw7zPIPA2FmWAqq/HQmPQCdJh4W/8mnpkHwckLbv7GrgAdYOupDE6k5OPiqGPuqAiuHdyJPx4cy/s3D2FoZ28u7ebHjcPCLvy9NiM/N0c8nSuWfUT4yXREIYQQjRBiCdKzTkNRFlCpDZtUeBdCiA7D7kz6sWPHGDx4cLXtgwYN4tixY3ady8/PD51OR2pqapXtqampBAUF1XlsYWEhX331Fc8//3yd+xkMBgwGg13j6jDO7QOTEZw81UJvyQfVrPqZTVX3c3AGJw9w9Qf3YBSPECalF3K3PoEJugOU4ohhzkoIsD97/ZEli37D0DA8XdRAV6vVMLlvEJP71v1voLXQaDR0DXBjX1w2IEXjhBBCNJKLD/hEQtYZSDoAXS+3zdaSIF0IIToOuzPpBoOhWlANkJycjIODfTG/o6MjQ4YMYf369bZtZrOZ9evXM2rUqDqPXbVqFaWlpdx8s30tvUQlZ7er91GXw983w3374W//hKB+oK+UDS4vhoJUtS1MzFo0+z/jPu1qrtbtxqRouL9sAbn+Q+x++ePJeWw9lYFWA/Mv6dJEb6plWKcjgrRfE0IIcQFs/dL3A+qSKo0GsovKyCo01nGgEEKI9sLuTPoVV1zB448/zvfff4+npycAOTk5PPHEE0ycONHuASxatIhbb72VoUOHMnz4cJYsWUJhYaGt2vvcuXMJDQ1l8eLFVY775JNPmD59Or6+vna/prCIswTpEWPUe98oGPsP9QbqVPfSfCjJhdI8KEiHvHPExsaw8+BfdHEqZJ3uUn7PHcLM2Cwm9A6s+XVq8fHWWACu7Bvc5ivWWovHgVo4TgghhGiU0CHw1yrbunRnRx2hXs4kZhcTk1ZQrWtJSZnJrrotQgghWj+7g/RXX32VsWPH0rlzZwYNGgTAwYMHCQwM5H//+5/dA7jxxhtJT0/n6aefJiUlhYEDB/Lbb7/ZisnFx8ej1VZN+EdHR7Nt2zb++OMPu19PWJQbIWGP+rjzmJr30enVqXcuVT8QrC08zUvlJ7gqMhgPJz3siWfnmUy7gvTUvBJ+OHQOgDsubdtZdKgoHufqqMPfTZZXCCGEaCRbJn0fKApoNET5u9UYpO88ncltn+5h3pguPHblxSuYKoQQonnZHaSHhoZy+PBhli9fzqFDh3B2dmbevHnMnj27xp7pDbFw4UIWLlxY43ObNm2qtq1Hjx4oitKo1xIWyQfVaewuvuBv34Xdui6uW4Abkf5ufLknnl1nMu06x2c7zlJmUhja2ZtB4d52HdsaDeviQ5S/K2O6+kn7NSGEEI0X1A+0DlCYBrmJ4BVG1wA3Np9Mr1Lh3WRWeO7Ho5SWm/njaIoE6UII0Y7YHaSD2gf9rrvuauqxiIvJOtU9fBTYGVSesgTpXQPcGB6hfqN/LDmPnCIjXi6O9R5fWFrOF7viALhzbKRdr91aeTjpWf/w+JYehhBCiLZO7wyBfSD5kJpNtwTpULV43Df7EzmRorZDjc0spNhowtlRpr0LIUR70KggHdQq7/Hx8RiNVYuYXHPNNRc8KHERWIvGRVxi12GKohCTas2kuxPg4USUvyun0wvZHZvFpD71V2RftTeBvJJyInxdmNDLvnXsQgghRLsXOqQiSO8zvVqF92Kjidf+iLbtrigQnZrPwDCvlhitEEKIJmZ3kH7mzBlmzJjBX3/9hUajsU07t07xNZlMTTtC0fTMJojfpT7uPNquQ9PyS8kvLUerqegHPirKl9Pphew8ndmgIP27A+pa9HljuqDTytRwIYRorRISEtBoNHTq1AmAPXv2sGLFCnr37i0z6ppT6BDYuxRit0D8LnooOqI05yjJcaS4tJxPtseSmldKqJczod7O7InN4nhyngTpQgjRTtjdgu2BBx6gS5cupKWl4eLiwtGjR9myZQtDhw6tcf24aIVSDoMxHwyeENjXrkNPWbLoEb6uGBzUaXUjI9UK+w1Zl55bXMZf53IBuKKPZNGFEKI1mzNnDhs3bgQgJSWFiRMnsmfPHp588kmef/75Fh5dO2YtHpd8EJZOwvPzCaw3PMJ2pwco+GIO728+A8D/Te5hC8xPJOe1zFiFEEI0ObuD9J07d/L888/j5+eHVqtFq9VyySWXsHjxYu6///7mGKNoanE71PvwkaC1b/3aqTR1/VvllmPWIP1ESn69PVx3n8nErECknyvBns52vbYQQoiL68iRIwwfPhyAr7/+mr59+7Jjxw6WL1/OsmXLWnZw7Zl/TxgyDwJ6g3cXcA+mQKNed70T1lFWWkTfUA+m9g+hZ5A7AMeT81tyxEIIIZqQ3UG6yWTC3V29IPj5+ZGUlARA586diY6OrutQ0VpY16PbOdUdKtbDVQ7S/dwMdA9Uf95dTzZ9x2n1+VFR0t9eCCFau7KyMgwGta3kunXrbHVnevbsSXJycksOrX3TaGDqErh3JzxwEB4+wYt9fyFF8cYBEwM0p3liSi+0Wg29gj0AOJ6SJ51vhBCinbA7SO/bty+HDh0CYMSIEbzyyits376d559/nsjI9lGpu10zmyHekkm3s2gcVFR27xboVmV7Q6e87zidAcCYrn52v7YQQoiLq0+fPrz//vts3bqVtWvXMnnyZACSkpLw9ZUvWy+mqAB39pm7AXBDYBKjo9TraJS/G3qdhvyScs7lFLfkEIUQQjQRu4P0p556CrPZDMDzzz9PbGwsl156Kb/88gtvvvlmkw9QNLH041CcDXpXCB5g9+EVPdLdq2wfZQnSd9YRpKfll3AytQCNpmJ/IYQQrde///1vPvjgA8aPH8/s2bMZMEC9bvzwww+2afDi4uga4MZ+c3cAJnvG27Y7Omht1d9lyrsQQrQPdld3nzRpku1x165dOXHiBFlZWXh7e9sqvItWzLoePWw46PR2HZpZUGpbcx7p71rluRGWoPtkagEZBaX4uRmqHb/TMtW9d7AH3q7191MXQgjRssaPH09GRgZ5eXl4e3vbtt911124uLi04Mg6ntFRfuyNGgPxX+CWtk/tu2b53NU72IMTKfkcT85jYm8pyiqEEG2dXZn0srIyHBwcOHLkSJXtPj4+EqC3FWe3qfedx9h9qDWL3snbGRfHqt/v+Lg62orX7D6TVePx22PUqe6jZT26EEK0CcXFxZSWltoC9Li4OJYsWUJ0dDQBAQEtPLqOxdFByz/mXg8OTlCcBZkxtues69JPpEiFdyGEaA/sCtL1ej3h4eHSC72tUpSKTHpjisalW6e6u9X4vHVd+qbotBqftxaNGy3r0YUQok2YNm0an3/+OQA5OTmMGDGC1157jenTp/Pee++18Og6IAdHCBmsPk7YbdvcM1gqvAshRHti95r0J598kieeeIKsrJqzpaIVy4yBwjTQGSp6sNrB2iO9ay1B+pR+wQB8fzCJtLySKs/FZxaRmF2Mg1bD8Agfu19bCCHExbd//34uvfRSAFavXk1gYCBxcXF8/vnnUoempYRZagFUCtKtmfSzmYUUGctbYlRCCCGakN1B+ttvv82WLVsICQmhR48eDB48uMpNtGJxltZrnYaC3snuw2srGmc1LMKboZ29MZrMfLwttspz2y1V3QeGeeFqsLsUghBCiBZQVFRka7v6xx9/cO2116LVahk5ciRxcXENOseWLVuYOnUqISEhaDQa1qxZU+f+mzZtQqPRVLulpKRc6NtpH8JHqvfxFUG6n5sBPzcDigLRKZJNF0KIts7uaGn69OnNMAzRbMqNkHRADdAPrlC3NWI9OlTqkR5YcyZdo9Gw4LKuzFv2J1/siuPe8VF4uagF4mSquxBCtD1du3ZlzZo1zJgxg99//52HHnoIgLS0NDw8PBp0jsLCQgYMGMDtt9/Otdde2+DXjo6OrvIasgbeopMlk54RDUVZ4KLOTusV7M7WU6UcT85nULh3HScQQgjR2tkdpD/zzDPNMQ7R1M7th3XPQMKfUH5e39Tuk2o+pg55JWWkWKaw1zbdHWB8D396BXtwPDmPZTvO8uCE7iiKwk5rf3QpGieEEG3G008/zZw5c3jooYf429/+xqhRowA1qz5o0KAGnePKK6/kyiuvtPu1AwIC8PLysvu4ds/VF3y7qkvYEvdC9ysAtcL71lMZHE+W4nFCCNHW2T3dXbQRG16A2C1qgO7iC72mwuSX4d7d6nR3O1mz6IEeBjycam/dpmbTowBYtuMshaXlRKfmk1FgxEmvlW/3hRCiDbn++uuJj49n7969/P7777btl19+OW+88UazvvbAgQMJDg5m4sSJbN++vd79S0tLycvLq3Jrt8JGqPcJu2ybrMXjaqrwvvVUOjd/vJstJ9MvyvCEEEJcGLsz6Vqtts52a1L5vRUwFla0Wpv7A3QZa+ul2lj1rUev7Mq+wXTxO0lsRiFf7olHa3ntYRE+ODrI90JCCNGWBAUFERQURGJiIgCdOnVi+PDhzfZ6wcHBvP/++wwdOpTS0lI+/vhjxo8fz+7du+usfbN48WKee+65ZhtXqxI2Ag4uh4Q9tk22NmzJ+SiKYvusVlJm4h+rDpGaV8qO0xk8MaUX8y/pIq1zhRCiFbM7SP/uu++q/FxWVsaBAwf47LPPOs7FsbWL3QKmUvAKb5IAHSqtR69jqruVTqvh7nGRPPrNX3y45Qw9LP3Tx8h6dCGEaFPMZjMvvPACr732GgUF6nXA3d2dhx9+mCeffBKttum/eO3Rowc9evSw/Tx69GhOnz7NG2+8wf/+979aj3v88cdZtGiR7ee8vDzCwsKafHytgjWTfm4fmMpApyfK3w29TkN+aTmJ2cWE+bgA8MWuOFLzSnHUaTGazLzw83GOJuWx+Np+OOl1LfgmhBBC1MbuIH3atGnVtl1//fX06dOHlStXMn/+/CYZmLgAJy1TErtNapIAHeBUqlottiFBOsCMQZ1Ysu4UybklpOWXAjAmSoJ0IYRoS5588kk++eQTXn75ZcaMUYuObtu2jWeffZaSkhJefPHFizKO4cOHs23btjr3MRgMGAyGizKeFufXHZw8oSQXUv6C0MHodVq6BrhzPDmP48l5hPm4UFhaznubTgPw/LQ+lJabef6nY3x34Byn0wv44JYhBHs6t/CbEUIIcb4m+wp85MiRrF+/vqlOJxpLUeDUWvVxtyua7LQx6dbp7g0L0h0dtNx5aaTtZw8nB3qHNKwSsBBCiNbhs88+4+OPP+aee+6hf//+9O/fn3vvvZePPvqIZcuWXbRxHDx4kODg4Iv2eq2eVltpXXrlKe/qzLXjyeoX68t2nCWz0EiErwvXDenEraMj+N/tw/F20XM4MZepb23nXE5xtdMLIYRoWU0SpBcXF/Pmm28SGhraFKcTFyLtGOQlgoMTRFzSJKcsNppIzFYv4g3NpAPMGh6Gj6vagm1UlC86rax/E0KItiQrK4uePXtW296zZ0+ysrIadI6CggIOHjzIwYMHAYiNjeXgwYPEx8cD6jT1uXPn2vZfsmQJ33//PTExMRw5coQHH3yQDRs2sGDBggt/Q+1JmKUuQEJFv/TelnXpx5PzyC0u44PNahb9wQnd0evUj3yju/rxw8JL6BrgRkZBKe9bMu1CCCFaD7uDdG9vb3x8fGw3b29v3N3dWbp0Kf/5z3+aY4zCHqf+UO+7jAVHlyY55en0AhQFvF30+Lo1fCqhi6MD/7iiB1oNzBzSTtcFCiFEOzZgwADefvvtatvffvtt+vfv36Bz7N27l0GDBtlati1atIhBgwbx9NNPA5CcnGwL2AGMRiMPP/ww/fr1Y9y4cRw6dIh169Zx+eWXN8E7akdqyKT3DLIUj0vJ45OtZ8grKadbgBtTB4RUPdTHheen9QFg1b4EsgqNF2fMQgghGsTuNelvvPFGlYqgWq0Wf39/RowYgbe3tNdqcc0w1f2kZT16t8D6K7ufb86IcG4cFiZZdCGEaINeeeUVrrrqKtatW2frkb5z504SEhL45ZdfGnSO8ePHoyhKrc+fP23+//7v//i///u/Ro+5wwgdAhqdOnsuNxE8O9mmu5/NLOKTbbEAPHxF9xqvwaMifekX6slf53L5fOdZHpzQ/aIOXwghRO3sDtJvu+22ZhiGaBLF2RBv6ZnabWKTnfZkqroevXtgw6e6VyYBuhBCtE3jxo3j5MmTvPPOO5w4cQKAa6+9lrvuuosXXniBSy+9tIVH2IE5ukJQP0g+qE559+yEr5uBAHcDafmlFBpN9AnxYFKfoBoP12g03DU2kvu+PMBnO87y97FRODtKtXchhGgN7A7SP/30U9zc3Jg5c2aV7atWraKoqIhbb721yQYn7HR6Iygm8OsB3hFNdlprZffujcikCyGEaNtCQkKqVXE/dOgQn3zyCR9++GELjUoA6pT35IOw/nk4sBx0et51yCdOr+Wd8mn844phdfZDv7JvEJ28nUnMLmb1vgRuGRVx0YYuhBCidnavSV+8eDF+ftVbaQUEBPDSSy81yaBEI1nXo3dvuqnuACfTLNPdAyRIF0IIIVqNqL+p99ln4fR6OPkbQ4u3c51uK894/sz4Hv51Hu6gq+jE8vG2WEzm2pclCCGEuHjszqTHx8fTpUuXats7d+5cpfCLuMjM5mZZj15kLCchS63s3tjp7kIIIYRoBt0nwa0/QkEamMrAXEZRcjQuf77NGP2pOrPoVjOHduKNdSeJyyzi96MpTOknre6EEKKl2Z1JDwgI4PDhw9W2Hzp0CF9f3yYZlGiEpANQlAGO7hA+qslOezqtEABfV0e7KrsLIYQQoplpNGo3l37Xw8DZMHguLhMeA40Wh/xEyD1X7ylcHB2YO7IzAB9sPl1nkT8hhBAXh92Z9NmzZ3P//ffj7u7O2LFjAdi8eTMPPPAAs2bNavIBigayTnWPugx0+iY7bUVld8miCyFER3HttdfW+XxOTs7FGYiwn8EdAvtCymFI2AWe19V7yNzREXyw5QyHEnPZHZvFyEhJugghREuyO0j/17/+xdmzZ7n88stxcFAPN5vNzJ07V9akt6RTv6v33Sc16Wmt69GlaJwQQnQcnp6e9T4/d+7cizQaYbfwUWqQHr8b+tYfpPu5Gbh+SCeW747nwy1nJEgXQogWZneQ7ujoyMqVK3nhhRc4ePAgzs7O9OvXj86dOzfH+ERDFKSp090BujZd6zWAU5b2a90CJJMuhBAdxaefftrSQxAXInwE7PkA4nc2+JA7Lo1kxZ54NpxI48dDSUwdENKMAxRCCFEXu4N0q27dutGtW7emHIuwR2EmJO6BhD1wZqO6LXgguAfadZqSMhPlZgU3Q83/FCqmu0smXQghhGgTwkaq96lHoDRfnQJfjy5+rtw2OoJPt5/lwZUHcXTQ1tpjXQghRPOyu3Dcddddx7///e9q21955ZVqvdNFM8g+C++Ogv9EwpezYNvrFVn0AfbVBDCbFWZ9uIvRi9eTll9S7fnC0nISs62V3SVIF0IIIdoEz1DwDAfFDIl/Nviwf17Vm2sHh2IyKyxcsZ+N0WnNOEghhBC1sTtI37JlC1OmTKm2/corr2TLli1NMihRhz0fQdox9bFfDxh0M1zzFizYAyPvsetUfxxL4WBCDnkl5Ww4Xv1CHJOmTnX3c3PEx9XxgocuhBBCiIsk3JJNj9/d4EO0Wg2vXNefq/oHU2ZSuPt/+9gRk9FMAxRCCFEbu4P0goICHB2rB2x6vZ68vLwmGZSohakcDn+tPp61AhbugWnvwOC54N/DrlMpisK7m07bft56qvpF2DbVPUCy6EIIIUSbEj5CvbdjXTqAg07LkhsHMqFXIKXlZuZ/tpe9Z7OaYYBCCCFqY3eQ3q9fP1auXFlt+1dffUXv3r2bZFCiFqc3QGEauPhBtysu6FTbYzI5nJiLRqP+vC0mA5O5am/UU5ZMendpvyaEEEK0LeGj1PvEveqX/HbQ67S8c9Mgxnb3p7jMxLxlf5KUU9wMgxRCCFETu4P0f/7zn/zrX//i1ltv5bPPPuOzzz5j7ty5vPDCC/zzn/+0ewDvvPMOERERODk5MWLECPbs2VPn/jk5OSxYsIDg4GAMBgPdu3fnl19+sft126RDX6r3/WZecC/0dzfFAHDTiHA8nBzILS7jUGJOlX2kaJwQQgjRRvn3AoMnlBWqBeTsZHDQ8cHNQxgQ5kV+STmPffsXiqLUf6AQQogLZneQPnXqVNasWUNMTAz33nsvDz/8MOfOnWPDhg107drVrnOtXLmSRYsW8cwzz7B//34GDBjApEmTSEuruVCJ0Whk4sSJnD17ltWrVxMdHc1HH31EaGiovW+j7SnOgRM/q4/tLBB3voMJOew4nYmDVsM947tySTc/ALaerDrl3dp+TYrGCSGEEG2MVgthw9XH8bsadQpnRx2vzRyAo4OWLSfT+XpvQhMOUAghRG3sDtIBrrrqKrZv305hYSFnzpzhhhtu4B//+AcDBgyw6zyvv/46d955J/PmzaN37968//77uLi4sHTp0hr3X7p0KVlZWaxZs4YxY8YQERHBuHHj7H7dNunYGjCVQkBvCL6w9/vuRjWLPm1gKKFezlzazR+ALafSbfsUlpZzzjK1TXqkCyGEEG2QdV16QuOCdICuAW7844ruALzw03HbZwMhhBDNp1FBOqhV3m+99VZCQkJ47bXX+Nvf/sauXQ2/CBiNRvbt28eECRMqBqPVMmHCBHburLnIyQ8//MCoUaNYsGABgYGB9O3bl5deegmTyVTr65SWlpKXl1fl1iYd+kq9HzAL20LyRjiVms8fx1LRaOCe8ZEAXGrJpB9MyCG3uEzdz1bZ3YC3VHYXQggh2h7ruvT4XXABU9XnXxLJ4HAv8kvLeeybw9WmvZvMCvvissgtKruQ0QohhLCwK0hPSUnh5Zdfplu3bsycORMPDw9KS0tZs2YNL7/8MsOGDWvwuTIyMjCZTAQGBlbZHhgYSEpKSo3HnDlzhtWrV2Mymfjll1/45z//yWuvvcYLL7xQ6+ssXrwYT09P2y0sLKzBY2w1ss6o1Vk1Wuh3wwWd6r3NakX3Sb2D6Gqp2t7J24VIf1dMZoWdp9Up79b16FI0TgghhGijQgaD1gHykyEnvtGn0Wk1/GfmAAwOWraeyuCrP9Vp7+UmM9/sS2TiG5u57r2d3P3FvqYauRBCdGgNDtKnTp1Kjx49OHz4MEuWLCEpKYm33nqrOcdWjdlsJiAggA8//JAhQ4Zw44038uSTT/L+++/Xeszjjz9Obm6u7ZaQ0AbXUx2yVNOPvAw8ght9moSsIr4/mATAvZdFVXlurGXK+2bLuvRTtiBd1qMLIYQQbZKjS8USuUauS7eK8nfjkUlqu9cXfjrGx1vP8LfXNvPwqkOcSS8EYOeZTNuX/EIIIRqvwUH6r7/+yvz583nuuee46qqr0Ol0F/TCfn5+6HQ6UlNTq2xPTU0lKCioxmOCg4Pp3r17ldfu1asXKSkpGI3GGo8xGAx4eHhUubUpilJR1X3A7EadIreojG/2JXLflwcwmRUu6epH/05eVfYZ192yLv1kOoqicNJSNK6bZNKFEEKItss65f0C1qVbzRvThaGdvSk0mnjh5+PEZxXh6+rIo5N7Mr6H+jniqz1tMBkihBCtTIOD9G3btpGfn8+QIUMYMWIEb7/9NhkZGfUfWAtHR0eGDBnC+vXrbdvMZjPr169n1KhRNR4zZswYYmJiMJvNtm0nT54kODgYR8c2vm7abIKv58Kyq+HYD2B9j/E7IScOHN2h51UNPl1puYmv/0zgtk/3MPTFtTy86hAHE3Jw1Gl5cEK3avuPiPTBUaflXE4xsRmFkkkXQggh2oMwS/G4+N0XfCrrtHdvFz3+7gaeuqoXWx+9jHvGR3HrqAgAvj2QSElZ7bWChBBC1M+hoTuOHDmSkSNHsmTJElauXMnSpUtZtGgRZrOZtWvXEhYWhru7fQHdokWLuPXWWxk6dCjDhw9nyZIlFBYWMm/ePADmzp1LaGgoixcvBuCee+7h7bff5oEHHuC+++7j1KlTvPTSS9x///12vW6rlHYcjn2vPj67FXy7wuj71SAdoM80ddpaA5SZzMxftpdtMRVfovQIdOfKfkFMHRBClH/17LiLowNDI7zZcTqTX4+kkJRbAkD3AAnShRBCiDYrfKR6n3ZMbefq5Aklueo6db0LeHe263Rd/FzZ+fjlOGg1OOgqcj1ju/sT7OlEcm4JfxxL5ZoBIU34JoQQomNpcJBu5erqyu23387tt99OdHQ0n3zyCS+//DKPPfYYEydO5IcffmjwuW688UbS09N5+umnSUlJYeDAgfz222+2YnLx8fFotRUXgLCwMH7//Xceeugh+vfvT2hoKA888ACPPvqovW+j9Uk9ot67Bqit1jJj4MdKXz40cKq7oig888NRtsVk4OKo4+5xUUzpF0zXBrRRG9vdnx2nM/lsx1kAAtwNeLro7X0nQgghhGgt3ALAJ1ItQvveGCjOhrLCiufHPQbjH7Orc4yTvvqSR51Ww8yhYby5/hRf7YmXIF0IIS6ARjm/j0YjmEwmfvzxR5YuXWpXkN4S8vLy8PT0JDc3t3WtT//jKdjxFgy/Cy5/GvYtg53vQn6SenFduA+09a9OWLotlud/OoZGAx/eMpSJvQPrPcbqaFIuV725zfbzJV39+OKOEY15N0IIIezQaq9NbZj8Tiv5+R/w50dVt1kz6gBD58OU/4D2wuoNJWYXcekrG1EU2PzIeDr7ul7Q+YQQoj2x57pkdya9JjqdjunTpzN9+vSmOF3HlGLJpAf2AYM7jL4Phv8dzmwE/x4NCtA3nEjlhZ+PAfDElb3sCtABegV54OdmIKOgFJCicUIIIUS7MOEZiLoMDB7gEQLuweoSuj0fwS+PwN5PoCgDrv0IHAyNfplO3i6M7ebP5pPprPwzgf+b3LMJ34QQQnQcdvVJF80o9ah6H9ivYpuDI3SfBN4R9R5+IiWP+1YcwKzArGFh3HFpF7uHoNVqGNvNz/ZzN1mPLoQQQrR9Bkvx2S6Xgm9URY2b4XfCzE9B56jWxVl+PZTkXdBLzRoWBsCqfYmUmcz17C2EEKImEqS3BgVpUJgGaCDA/m+d0/NLmb9sL4VGE6MifXl+Wl80dqwtq+zS7hVBenfJpAshhBDtW58ZcNNqcHSD2C3wwVhYswC2LYETv0BGjNqBpoEu7xWIn5sj6fmlbDiR1nzjFkKIdqxJpruLC2QtGucbBY72r996fW0053KKifRz5f2bh+Do0PjvXi7p6o9Oq0GrgW7Sfk0IIYRo/yLHwW0/wRfXQ3aseqvMvyfctRn0TvWeytFBy3VDOvHB5jOs/DOBSX2CmmnQQgjRfkmQ3hrYprr3sfvQtPwSvtl3DoCXr+t/wdXY/d0NfHzrUAA8naWyuxBCCNEhhAyChX9C7GZIPwkZllv6CfW2/3MYcVeDTjVrWDgfbD7Dpug0knOLCfZ0bubBCyFE+yJBemtgKxrX1+5Dl20/i9FkZnC4F8MivJtkOJf1CGiS8wghhBCiDXHxUae/V/bnx/Dzw7DtDRhya4MKy3Xxc2VkpA+7zmSxam8i91/erZkGLIQQ7ZOsSW8NbJl0+4L0gtJy/rcrDoC7x0U1eh26EEIIIUSNBt0CHqFqS9j9nzf4sJlD1AJyPx9Obq6RCSFEuyVBeksrN6rTyMDu6e5f7Yknv6ScSH9XJvSyr92aEEIIIUS9HAxwyUPq421vQHlpgw6b0CsQB62G6NR8YjMKm3GAQgjR/kiQ3tIyT4G5TO1d6hXe4MOM5WY+3qoWdvn72Ei0WsmiCyGEEKIZDLpF7a2edw4OfNGgQzxd9IyK8gXg96MpzTk6IYRodyRIb2mVi8bZMV39h0NJpOSV4O9uYPqg0GYanBBCCCE6PL3Tedl0Y4MOu8JS2V2CdCGEsI8E6S0t5S/13o716GazwodbTgNw+5guGBx0zTEyIYQQQgjV4FvBLQhyE+Dg8gYdMql3IBoNHIjPISW3pJkHKIQQ7YcE6S2tEe3XNp1M42RqAW4GB+aMaPgUeSGEEEKIRtE7wSUPqo+3vt6gbHqAhxODwrwAWHtMsulCCNFQEqS3tFRL+7Wgfg0+5P3NZwCYMyJcepkLIYQQ4uIYchu4BUJuPGxaDH+thj8/UafAb3ypYnZgJZNsU95TL/JghRCi7ZI+6S2pIB0KUgEN+Pds0CEH4rPZE5uFXqfh9jFdmnd8QgghhBBWemcY8wD8/gRse7368wdXwH37qvRSn9QniMW/nmDXmUxyiox4uThWOcRkVkjJKyHUy7m5Ry+EEG2GZNJbUpplqrtPFzC4NeiQz3acBeCaAaEEeTo108CEEEIIIWow9HboeTWEDIIuY9XHA+aAq7+6Xn3fsiq7R/i50iPQnXKzwvrjaVWeM5sV7vjsT8a8vIFdZzIv4psQQojWTTLpLSnFMtW9gUXj0vNL+fmvZABuGx3RTIMSQgghhKiF3hlm1VA4bu9S+Okh2PIfGHQzOLranprUJ5Do1Hx+P5rCdUM62ba/t/n/27vv8Cir9P/j7ymZSe89gYQOAUIHAxYUVkRWUdRFFxX7WvCroru2te4q/lZXLOuia2N3dUVRwYKKgBQRBCmB0EINoaQCqaTOPL8/JhmICRAgySTweV3XXEzmKXOeDMnJ/Zz73GcHC9PzAPh+Yw7ndAxr9uaLiLQFGkn3JHfRuMYF6TNWZlLlMOjXPpje8UHN2DARERGRk9DvBghJhNI8+HlanU2jernmpS/ZlkdZpQOAVRkHeXneVvc+v2QcbLGmioi0dgrSPSmnpsBK9ImD9CqHkw9XZAJwY0pCc7ZKRERE5ORYvODCx13Pf3oNyg65NyXFBBIf4kN5lZPFW/MoOFzJ/320FofT4Lwu4QBs3F9ISUW1J1ouItLqKEj3FEcV5KW7njdi+bV5m3LILionzM/Gpb1jmrlxIiIiIiep19UQ2RMqCl2Beg2TyeSu8v7dhiwemrme/YXlJIb5Mu36AcSH+OA0YM3uQ8c6s4jIWUVBuqfkbwNHJdgCIPjEI+P/WZ4BwHWD22O3Wpq5cSIiIiInyWyGi/7ser7iTSg+suxabZD+xbr9zN+cg81i5h+/74+/3crgxFBAKe8iIrUUpHuKez56TzCZjrtrenYxP+88iNnkWhtdREREpFXqNhriB0HVYfjx7+6XB0QY/M53NZPMn9PftJXHRnejV5yrvs6gDq4gfeUuBekiIqDq7i1jyUuw/hPodgn0vR4iukJObWX3E6e6146iX5wUTazWERUREZHWymSCEU/Cvy9zVXy3eEHGUixZ6/gbBnjBg3yKkfoRmG+EPtcyqGYkfceevVTu+BHbgS1g9XZViTeZMAyDm6f/Ql5xBR/cOoQQP9sJGiEi0rYpSG9uTicsew3KCyE/HX561XWHuazAtf0EReOKyquYtXYfADcOVcE4ERERaeU6nA8dh8PORbD8H+6XHeHdyfWKIzpvGab8dPj+cZj/NJ3iB/Kz91aiOQD/Peo8Nj/oNY7Vuw+xqGaptj9+up63bxyA6QRZiCIibZnS3ZtbfrorQPfyha6jwWSBvb/AgW2u7SdYfu2z1Xs5XOmgS6Q/KVo/VERE2pglS5Zw2WWXERsbi8lkYvbs2Sc8ZtGiRfTv3x+73U7nzp2ZPn16s7dTmtjoFyHhXOg7Aca9DQ+mY5m0gpg/fI7poXT47VSI7Q/OKkyZy10BOlBkj3EVnwP48WUwDD5bs8992vmbc5i+LMMDFyQi0nI0kt7c9qxw/Rs3AH4/w1VEJe0TWP8xeAdDTN9jHup0Gvx3+W7Ateya7hqLiEhbU1paSp8+fbjlllsYN27cCffftWsXY8aM4c477+TDDz9kwYIF3HbbbcTExDBq1KgWaLE0iYiucPOchrd5B8HAW1yP7A2QtY6v9vnw2FIHAxMSeH98Z3ilN+SkUbn5W75e7yqY+9vkGL5en8WUb7YwKDHUPaddRORMoyC9ue1Z6fq33RDXvwFRMPRe1+MEXvthGzvzS/G3W7myf3wzNlJERKR5jB49mtGjRzd6/zfffJMOHTrw97+7io716NGDpUuXMnXqVAXpZ6LoXhDdi8SIQoqXLmXV7kM4vEOwDLoVfnqV0nlTKC5/mLhgX169th+V1U6+35TDpP+t4ev/Ow9/e8v+KfvPRdvJPHCYv17RC6tFCaki0jz026W51Y6k1wbpjfThit28Mt+VEv/I6O4t3gmJiIh4wvLlyxk5cmSd10aNGsXy5cuPe1xFRQVFRUV1HtJ29IgJwM9mobi8mvTsYkiZBFZvQg6tZ6h5I1f2i8NiNvG3q5OJC/Yh48BhHp+VhmEYLdbGfQVlvDg3nRm/7OHnnapELyLNR0F6cyo9AAe2u57HD2z0YXM3ZvPEbFf193sv6sz156hgnIiInB2ys7OJioqq81pUVBRFRUWUlZUd87gpU6YQFBTkfrRr1665mypNyGox0z8hBKhZL90/ksO9fg/AJMtsxvWPAyDY18Zr1/XFYjbxRep+nvlqE6/M38pzczbx6OdpTP44lSVb85qljbPW7KX2nsDirbnN8h4iIqAgvXnVjqJHdAff0EYdsmLnAe79aC1OA64d1I7Jv+najA0UERE5Mzz66KMUFha6H3v27PF0k+QkDa5Zim1lhmuU+gu/q6kyLAy1bKJj2Ub3fgMSQt1/H01flsEr87fx9o+7+GhlJp+v3cdTX26sf/LTZPyqgF1ttXkRkeagHOrm5E51H9yo3bdkF3Hbf1ZRWe1kZI8o/npFLxWLExGRs0p0dDQ5OTl1XsvJySEwMBAfH59jHme327Hb7c3dPGlGgzrUBOm7DmIYBv/e6MDsOI/x1kXw40swYaZ737su6ERFlYPdBw/jZ7fib7fiZTHxxsIdZBwopazSgY/N0mRtW5N5iF35pfh4WaiodrAtt4R9BWXEBR/7/6SIyKlSkN6cfl007jiqHE5unb6K4vJqBiaE8Pp1/VSQREREzjopKSl88803dV6bN28eKSkpHmqRtJS+7YLxspjIK67guw3ZbMku5h3LWH5nWoJp2/eQtQ5i+gBgNpuYfHG3OscbhsH/VmRy6HAVO/JKmrT6+6erXaPoo3tHs/vAYVbvPsSSrXlcN7h9k72HiEgtRYHNpboS9q9xPW9EkL77wGH2FZTh42XhnYkDm/Tur4iIiKeUlJSQmppKamoq4FpiLTU1lczMTMCVpn7jjTe697/zzjvZuXMnf/rTn9iyZQv//Oc/+eSTT3jggQc80XxpQd5eFvrEBwPw7NebAOiS1AdTz5ql+75/AjbOgu0LYO8qyNsKTof7eJPJRJeoAAC25hQ3WbvKqxx8vX4/AFf3j+eCrhEALErXvHQRaR4aSW8u2euhuhx8QiGs8wl3zy0qByA22JtgX1tzt05ERKRFrFq1igsvvND99eTJkwGYOHEi06dPJysryx2wA3To0IE5c+bwwAMP8OqrrxIfH88777yj5dfOEoM6hLJq9yGyCl1/F43rFw9hD8KGT2HXYtfjaDF94KZvwO4PQNcof1buOsjWnJJjvodhGGQcOExCqC9m84mnFX6/KYfi8mrign04p2MY/t5WXp63lZ+2H6DK4cRLmY8i0sQUpDeXo5dea8S88uyaID06yLs5WyUiItKihg8fftxlsqZPn97gMWvXrm3GVklrNTgxlGnsACDMz8YF3SLAEgWXvgTb50N5EVQUuf4tyXGlwM97An47FYCuNSPp23OPPZL+0co9PDYrjacvS+KmYR1O2KbPVu8FYFz/OMxmE71igwj1s3GwtJLVuw9xTsew071sEZE6dOuvuZxk0bjaID0qUEG6iIiInJ36J4S4xzYu7xt7ZJR68O3w+4/hlm/hrp/ggTSY8Ilr26r3YNs8ADpHukbUjzeSvmCzqzDhdxuzT9ienKJyftzmquQ+rn884JoPf36XcAAWN9NybyJydlOQ3hwM46SKxgHk1KR1RStIFxERkbNUkI8XwzqFY7OaT1yUreNwGHKX6/kX98Dhg+6R9D2HDlNW6ah3iGEYrNtbCEDqngKqHM7jvsWstftwGjAwIYQO4X7u14d3iwRgsZZiE5Fm0CqC9DfeeIPExES8vb0ZMmQIK1euPOa+06dPx2Qy1Xl4e7eywLZwDxRngdkKsf0adYjS3UVERETgrRsGsPiPw90B93GNfArCu7lS37++n3A/G6F+NgwDtufWH03PLionv6QCgPIqJ5v2Fx3z1IZhuFPdrxoQX2fbeV3CMZlgU1aRu66QiEhT8XiQ/vHHHzN58mSeeuop1qxZQ58+fRg1ahS5uceumBkYGEhWVpb7sXv37hZscSNk1qS6x/QBm2+jDskucnUYSncXERGRs5mf3UpMUCPXH/fygXFvuQZGNn0B6z+hizvlvf689HV7Cut8/UvGwWOeOm1fIdtyS7BbzYxJjqmzLczfTnLNEm9KeReRpubxIP3ll1/m9ttv5+abbyYpKYk333wTX19f3nvvvWMeYzKZiI6Odj+ioqJasMWNcHTRuEZSuruIiIjIKYjtBxc84nr+zR8ZGnQQG1VsayBIT9tXAICXxTXxffXuQ8c87ac1o+ijekYT6O1Vb3vtUmwK0kWkqXk0SK+srGT16tWMHDnS/ZrZbGbkyJEsX778mMeVlJSQkJBAu3btGDt2LBs3bjzmvhUVFRQVFdV5NLuTLBrncBrk1aReKd1dRERE5CSd+wDED4KKQu7b8nu2ek/kTyuHwvNx8PfuMPdxqChhfc189DG9XSPjq3YfanD1AYfT4Ju0LMBV1b0hF3RzBek/bsun+gRz20VEToZHg/T8/HwcDke9kfCoqCiysxuuuNmtWzfee+89vvjiCz744AOcTidDhw5l7969De4/ZcoUgoKC3I927do1+XXUUVECORtcz+MbF6Tnl1TgcBpYzCbC/e3N2DgRERGRM5DFCle+5ZqfXsOMEypLXHWClv8D459DCN7zAwDXn5OAl8VEXnEFmQcP1zvd2sxD5JdUEuBtZWin8Abfsk98MEE+XhSWVbmL0Z2MzAOHeWjmuuPOixeRs5PH091PVkpKCjfeeCN9+/blggsu4PPPPyciIoK33nqrwf0fffRRCgsL3Y89e/Y0bwP3rQbDCUHtIKjhO6+/ll2T6h7hb8diPvGa6iIiIiLyK2GdYNJKDjywj97l7zCo/A3K7loF4z+E4PaYCvfyOi8wzfYqyUHl9K6ZU74qo37K+/ebXMu0jegeic3a8J/LVouZc2uXYks/di2lhlQ7nEz6aA2frt7LC99tOaljReTMZ/Xkm4eHh2OxWMjJyanzek5ODtHR0Y06h5eXF/369WP79u0Nbrfb7djtLTQ6XVYAGz93PT+J+ejuNdKV6i4iIiJyWsKC/LH5BZNXWsm2qgiSe3SBTheyfeafSdw6ndHmFTBtEK97JbDDy0zwkgjY3941wDL4dgx7IHNr1lC/uOfx/x69oGsEc9ZnsXhrHpMv7nbcfY/21pKd7tT7ZdvzKTxcRZBv/XnvInJ28uhIus1mY8CAASxYsMD9mtPpZMGCBaSkpDTqHA6Hg7S0NGJiYk68c1MzDMjZBEunwnuj4W8dYfV017b25zT6NDm1y68FKtVdRERE5HR1iXJVeN+WU7MMm82Pj4Nv5/LKv7LXtwdUlhBXupHzLWkkF/7g+vvth7/AB1exbU82uw8cxmY1u4vDHcvwmu3r9xWSV1zRqLZtyS7ilflbAfC1Wah2GszfnHPM/X/YksODn6yjqLyqUecXkbbP4+nukydP5u233+bf//43mzdv5q677qK0tJSbb74ZgBtvvJFHH33Uvf+zzz7L999/z86dO1mzZg3XX389u3fv5rbbbmv5xs+6E6alwPynIXMZGA7XXKhzH4C+v2/0abJV2V1ERESkyXSJdK2xvjX3SIX3dXsL2WQksvzCGXDrPIqu+DcPVt7J01U3Uj70j+ATAnt/wf/Ta/GhnPM6h+NnP37SaWSgN8nxQRgGLNxy4pT3KoeTh2auo8phMLJHFLef1xGAbzdkHXP/hz9L47M1e5mxMrOxly8ibZxH090Bxo8fT15eHk8++STZ2dn07duX7777zl1MLjMzE7P5yL2EQ4cOcfvtt5OdnU1ISAgDBgxg2bJlJCUltXzj4wbAptnQ4XzocjF0+Q2EJJ70aZTuLiIiItJ0uv5qJN3hNNi4z5VentwuDKITCWwHa+cHszO/lPPbD+SiXmPg32OJLUrlHa+/k9V9eqPea2SPKNbvLWTe5hx+N+j4BYrfXLSDDfuKCPLx4vkre3HwcCWvLtjGkm35lFRU4/+rmwLzN+W4R+jnbszhjvM7ncy3QUTaKI8H6QCTJk1i0qRJDW5btGhRna+nTp3K1KlTW6BVjdD399D/BvDyOa3THEl3V5AuIiIicrq6RNWMpNeslb4zr4TSSge+NgudI/3d+w1ICGFnfimrMg5x0SX9yLvif/jMuIphlo1UbvkTDPwYrEdNRywvArMVbL7ul0b2iOLleVv5cVse5VUOvL0sDbZpc1YRr/2wDYBnx/YkMtCbiAA7HcP92Jlfyg9bcrm8T2ydYz5ccWT0fE3mIXKLy4kM0N+LImc6j6e7t2l2/9MO0EHp7iIiIiJNqWtNkL73UBmlFdXuJdJ6xQbVWUlnYGIIcKTC+5yDcdxc+SfKTXZsGQth+m/hf9fCtHPhhfbwQjuYmgR56e5z9IgJIC7Yh/IqJz9tz2+wPUenuV+cFOUOxk0mE5f0chWn+zatbsr7rvxSlm7Px2SC9qG+GAbM23TsuetNqbzKwZz1WVRUO1rk/USkLgXprUBOkSuNSenuIiIiIqcv1M9GuL8NgB15JaTtLQCgd3xQnf0GJoYCsG5vAZXVTr7flMMvRncW9H0NrN6wdyVs/RZy0qC8Zi30skPwv/Fw+CDgCrRH9ogEOGYBuE9W7WHj/iJCfL147sremExHbhRc2ttV/HhReh6HK6vdr39UMwd9eNcIxtek0c/d2DJB+t+/T+ee/63hnwt3tMj7iUhdCtI9rKSimpIK1y9kjaSLiIiINA138bicEvdIevKvgvSO4X6E+tmoqHaNgq/Y5Qq8e597Odz6PQx/FMa8DBM+hbt/hvvWQ3B7OLQLZt4EDlfF9ZFJrlpK8zfn4nQadd7D6TR498ddANx7URciAuqu5tMzNpD4EB/KqhwsTs8DXCPZM1ftAWDCkARG1SwFt3xHfrNXeTcMg2/SXEvQLdjSMjcFRKQuBekeVpvqHmC3nrCCqIiIiIg0Tu0ybJv2F7EpqwiAPvHBdfYxmUz0b+9KeX9xbjoOp0H36ADah/lCTB8Y/ggMutVVHDiyB4QkwHUzwOYPuxbD3McAGNIhDH+7lbziCtbXFKirtWBLLjvzSwn0tjZYWM5kMrlH07/d4AqOv9uQzaHDVcQGeXNh90g6R/rTKcKPKofRqCryp2NrTgn7CsoA2Li/iIOllc36fiJSn4J0D8tRZXcRERGRJldbPO7bDVlUVjsJ9LaSEOZbb7/aeem1gfzFNaPWxxTVE8b9CzDByn/Bqvdda6p3c62ZPv9X88bf/nEnAL8fklCvenut2nnpCzbnUF7l4MMVuwG4dnB79xz62tH0uRuzj9++03R0yr5hwLIdDc+zF5HmoyDdw1Q0TkRERKTpda2p4p5V87dWcnxwnbngtQbVBOm1Lq5JXT+u7mPgoj+7nn/zEGz5htGdfQGjTpC7bk8BK3cdxGo2cdPQxGOerm98MDFB3pRWOnh36S5+yTiExWxyz0WHI0H6onRXFfnm8kPNSH3tnP6l2xSki7Q0Beke5l4jXUG6iIiISJOprfBe69dF42r1igvCZnX9SRwX7EPP2MDGvcF5D0Kvq8FZDTOu47ffDGGD/VbeOHgH5e+MgfWfuEfRL+8TS/RxsibNZpM7CH953lYAftMjqs7fh8nxQcQEeXO40tFsgfOBkgrWZLoq3T94cTcAftyWj2EYxztMRJqYgnQPc6+RHmQ/wZ4iIiIi0lghfjbC/Y/8fdXnGEG63WohOc617eKeUQ2OtjfIZIKx/4CeV4K363h/UzmdzFl4710Kn98OG2cDcNt5HU94utE1Ke+OmsJz15+T8Ku3M7lH+Zsr5X1Reh6GAUkxgYztG4uXxcS+gjIyDhxulvcTkYYpSPcwpbuLiIiINI8uNSnv4Ep3P5ZJF3XmvC7h3Hpuh5N7Ay8fuGY6PJIJj+3nk5TZXFf5OAt9LwHgRes0bmh/kKRGjM4PTAx131RIDPNlaKewevvUjrbP35xDtcN5cm1thNpq7iN6ROJrs7qL6i3dltfk7yUix6Yg3cNylO4uIiIi0iy61lR4D/e3EXOcdPPh3SL5761DiA+pX1iu0Wx+DBk4mOXOntxZcD1LjL74mCr5c/FfoGj/CQ+3mE1c1T8OgJuHdcBsrj+iP7hDKMG+Xhw6XMUvGYdOva0NqKx2smSrK41+RA/XiP15XcIBWLpd89JFWpKC9Bby6zUza2W7090VpIuIiIg0pdrR84EJoY1PYz8NCWF+dIn0p8Jp5p6KSWSY22Evy4GProXKE6eMP3hxN76cNIwbUxIa3G61mBnRvXlS3lfuOkhJRTXh/nZ3+v+5XVwV65ftONAsI/ci0jAtzN0CHpuVxpz1Wcz5v3Pr3KGtdjjJK64AlO4uIiIi0tSu6BeHwzA4vybYbAkjk6LYlltCMb5svOBfJK64HrLWwew74erpYDaD0wFVh11rnHkfSYW3Wc3HTcsHGNUzis/W7OXbDVl0CPejrMpBeZWD8ionnSP9GdcvrsFR+BOpTXW/qHuE+/jecUEE+XhRWFbF+n2F7vR3EWleCtJbwPxNORSWVfFtWja3n3+kcMiB0kqchiu9KcxfheNEREREmpLFbOJ3A9udeMcmNLpXNNMW7SAq0M7IYUMg8QP49+Ww6Qv4WyJUV0B1ec3eJhh4M1z6EpgtjTr/+V0j8PGykFNUwVNfbqy3/ZNVe/j7NX1oF9r41H3DMFiw2bX0Wm2qO7i+f0M7hfHthmyWbsuvF6R/tW4/q3cf4uFLuuNja1z7W8q7S3exPbeYZ8f2wsui5GFpW/Q/tplVVDvIrRkt//V8ntqicRH+diyncMdTRERERFqX5PhgPrh1CB/dfg52qwUShsLlr4HZCuWFRwXoAAaseg9mTnQF743g7WXhuSt7MaJ7JGN6xzCufxwThrTn+nPa42uzsHLXQUa9soT/rchs9NJpO/JKyDx4GJvFzLmdw+tsG1bz9a+XfVux8wD3f5zK9GUZTFu8o1Hv01Lyiit4bs4mPlq5h0XpKnonbY9G0ptZVsGRX8Qrdx2kotrh+oXNUWukaz66iIiIyBnj3C51A136/h46jYCyQ2DzBS8/17/bvofPboPNX8GH18C1H4I9oOGTHmVc31jG9QwGL19X+nyNO87rxEOfrmPlroM8NiuNuRuz+dvVyScsUDy/ZhQ9pVMYfva64UFt8bg1mYcorajGz24lr7iCez9a614u7q3FO/jdwPjTK7zXhL5ev5/aclBzN2bzm6So4x8g0spoJL2Z7Ssocz8vq3KwZneB+2v3GumBSnUXEREROaMFREFkdwhuD35hruXbksbChJlg84ddi11p8aUH6h9rGJC7BVa+DR9fDy92hClx8GwIPBcDL3aGV/vQfslkZtyUzJ/H9MBmNbN4ax43vrvyhCPqP7hT3SPrbUsI86NdqA/VToMVuw7gcBrcN2MtucUVdIn0Z3BiKBXVTqZ8u6VJvk1N4YvUI9X052/OoUpF76SN0Uh6M9t3qKzO1z9tzyelZt1LrZEuIiIicpbrOBwmfgkfXA3718C7IyGmr6uwXGWp61G4F0pzGz6+6rDrUZoHhzIwl+Rw23UzuKBrBJf9YynpOcVsyS6mR0zDa7UfKq1k1e6DAFzUvX6QDnBu5wg+WpnJj9vyWZtZwLIdB/C1WZh2fX8qqw1++/qPzFmfxY3nHGBIx/rru7ekjPxSUvcUYDaBv91KweEqVu466E7bF2kLNJLezPbWjKQHeLvuhxw9L13p7iIiIiJC3AC4ZS4ExsHBnbDxc9j6HWT86ArcS3PB6u0K6C96Am6dB4/uhT/ugPvWwV3L4Hf/caW/7/gBZt5El3Bv9/zyBZtzjvnWi7bm4jSge3TAMdPVa8/z+Zp9vP7DdgCmjOtN58gAkmIDuW5wewCe+WqTOwXeU2pH0Yd1Dmd0rxgAvtvQtMvViTQ3jaQ3s9qR9LF9Y/ng50zW7y2g8HAVQb5eR6W7K0gXEREROatFdIXbF7oCdACbnyvotvmDbyjE9AHrr6ZI2gPAr2aEOKoneAe75ranfwOf38GIbk8yf3Mu8zfnMumiLg2+7Zz1WQBcfJx520M7hWEyQWFZFQDXn9OesX3j3Nsn/6YrX63bz6asIj5ZtccdtLc0wzD4InUfAFf0jSPUz8bHq/Ywd2M2z1ze85SWphPxBI2kN7P9NSPpAxJC6BThh9OA5Ttdc42U7i4iIiIibgFRcM5drkf/G6H31dDtEmg3uH6A3pCOF8D4D8DsBRs/54o9L2DCybq9BeQVV4DTCRXFrjnuQOHhKhZvdVU/v6xP7DFPG+Jno3dcEADJ8UE88dukOtvD/O3cP7IrAC/NTXcH8y0tbV8hO/NL8fYyM6pXNEM7h+Fvt5JbXEHq3gKPtEnkVChIb2a1hePign3dqUI/1aS85xS5ltpQuruIiIiINImuF8PV74LJjM/GGfzo+zCLve4j8LXO8JcwmBIPU3vBtw+zZslXOBwOukcH0CXq+FXlH76kO2P7xjLt+gHulYqOdkNKAp0i/DhQWsnrC7Y119UdV22q+8geUfjbrditFvc8+7lKeZc2REF6M3I6DbIKa4L0EB93wYqftudTUlFNSUU1oJF0EREREWlCSWPhijcBE/HOfbQ352GvKgKjpsp50V5Y8SYX/nwzK+1387L3O7D4RVj2D/jlXUj9CDZ/DcVHAtthncN59dp+xAX7NPiWXhYzT17WE4DpyzLYuL+wua+yDofT4Kt1riD96FT8UT2jAfhuY3aj140X8TTNSW9GucUVVDkMLGYTUQF2AjqFYTGb2JlfyprdhwAIsFvrrUcpIiIiInJa+oyHuAFkbEtj8pcZVHoF8OkDl+Lt4wcZP1Ge9gXlG74i3FREePYXkP1Fw+eJ6A4dLoAO57vS7gGqy6G60vWvTzAExQNwQdcILukZzXcbs5n88Tq+vHdYg6PuALvyS9mcVURuUTl5JRXkFlVw6HAVQzqEct2Q9vif5N/Hy3ccILe4gmBfLy7oGuF+fXi3CGxWM7sPHD5ulXuR1kTRYTPaV3AYcI2UWy1mAi1m+sQHsSazgM/W7AWU6i4iIiIizSS8Mwlhndi3cAE5RRX8nGtleLcg6H4pMw8l8czqS7kucg9/6ZUD5QVQVXbkUZILORsgb4vrsfKtht/DZIEJn0DnkQA8d2UvVu0+SHpOMS/P28qjo3vUO2Tmqj386bP1NDSwPX9zDq//sI0bUhK4aWgHIgIaMRcfmF1TMO7S3jHYrEeShf3sVs7vEsH8zTl8tyFbQbq0CQrSm9HeQ0dS3Wud2zmcNZkFzN3oSh9SqruIiIiINBeTycRF3aP4aGUmP2zJZXg31xztr9ZlUY2VhEGXwnkdGz748EHIWAq7FsPOxXCgZq651ftIIbvyQvj0VvjDYghJJMzfzvNX9uaO/67mX0t2MrJHFIMSQ92n/HLdfh7+bD0Wo5qLI4sY6LOfrkYG7ap2EVqWQWZlIGsr49m0JJG7f0wkqe853D+6LyF+tmNeY3mVw73M2tgGCuBd0iua+ZtzmLsxmwd+0/Wkv4cHSyt5/6ddXNU/nsRwv5M+vq2odjh5cOY6/O1W/npFL0wmVcP3FAXpzehI0bgjQfqwzuG89sN2yqtcc4KiFKSLiIiISDMa0T2Sj1ZmsmBzLs9cbpBdVM7KjIMAjEmOOfaBvqGQdLnrAeCoArMVaoO36gp4fzTsWw0fX+9av93Lh4t7RnPNgHhmrt7L5E9S+fa+8/G3W/luQzYPfJzKeaZ1vOH7Fv5FBVBU9y17kkVPa/qRFzZA5QYvDG8/TF6+4OUD/pFw4ePQ4TwAftiSS0lFNbFB3nVuCNQa2SMSi9nEluxiMvJLTyrQNgyD+z9OZcnWPOZvzuXre8/FcoYu5fbDllx38b1rBrajb7tgzzboLKbCcc2odo30o4P0fu1D8LUdmZsTHdS4FB4RERERkVMxrHM4dquZfQVlbMkudq+NPjgxlJighgvBNcjidSRAB9do+u/+A77hkJ0GXz/gXt7tycuSiAv2Yc/BMp6bs5mFW3K596PV3Gr6kvdtL+LvKAB7ILRPgUG3w29fgZu/havfg3MfwOg8kkpv19xyG1WYygugeD8c3AGZy+E/Y2H5P6mudjD9pwwALu8b1+Ba6MG+NlI6hgG4s1kb64MVmSypWaZuc1YR/1ux+6SOb0s+WJHpfv7RUc+l5SlIb0a1a6Qfne5us5oZ0uHIHT6lu4uIiIhIc/KxWdxLAS/YnOOugn5Zn+OMojdWUDxc8z6YzLDuI/jlHQACvL146Zo+AHy0MpP/++9y/p/5nzzm9RFmnK514P+4A275Dsa8BANvhoSh0OsqGPk0pus/w/bIdj6/eClDy19jVNVL7Bj3DdwyF3r/DgwHzH2UtNfHsz4jC28vM+MHtTtmM0f1OlLlvbF25Zfy/JzNAAyu+fv9pe+3crC08pS+Va1Z5oHD7psR4JqWUFzumfXuRUF6s2oo3R1wL8UGSncXERERkeY3okcUAB+t3MO6vYWYTTC6dxME6eCq/D7yGdfz7x6F3cuhqpyUdj7cmRJFO1MO/7U8zTjLUgyTBS59CS57DazHnmde68qUXvRK6km6I5Y751dTHjMIxv0LLnkBJxb6Fc7jc9vTvH1ZBB3C/cDpcBW+Ky90pefXGJUUhckEazMLWFWT6n881Q4nkz9JpazKwdBOYXx42xCSYgIpLKvixbnpJzy+rflwpStD4PyuEXSO9KesysHsmtR3aXkK0puJYRhH0t1D6gbp53Y5EqRHq7q7iIiIiDSzi7q7CsbVDiIN6xxOuH8TTrscei8kXQHOKnj/EnguCp6P4ZG1I/jR/gB9zTsxfEIw3TALBt9eN23+OEwmEy9clUxEgJ1tuSW88O0WMJn4zOsyrqt4jDwjkCTzbs779jfwTAg8GwrPRcML7eHFTrDsdaiuJDLQm98NcI20P/3VRhzO46+Z/taSnazNLCDAbuXFa/rgZTHzzFjXOvAzfskkbW/LrgPfnCqqHcxc5Vp56voh7fn94PYA/G9FptaW9xAF6c2ksKyK0koHUH8kvVtUAEkxgYT52egY4e+J5omIiIjIWSQ6yJtecUeWH7ssuX4V9NNiMsHYf7jmmDckth+m2xdCxwtO+tShfjZ36vz0ZRn8/ft0Hv5sPSuMHnzc7wOIHwSG0/U4WnkhfP9n+OcQ2DKHP47qSoC3lQ37ivj4lz3HfL8N+wqZOm8rAE9f3tP9t/ygxFCu6BuLYcBTX27AeYJAv634Ni2bg6WVxAR5c1H3SMb1j8NmNbM5q4j1Z9DNiLZE1d2bSe3ya+H+Nry9LHW2mUwmPrkzBadh4G/XRyAiIiIizW9E9yg27CvCy2JiVM/opn8De4Cr+Ft5gWv9dJP5yMNqb/ToeUMu6BrBTUMTmb4sg9d/2A7A5X1iufvyvmCaB8VZrve0eLkeZits+AwWPAsHd8KM3xOeeB6v9B7LB6tzWPbdBi7374O/tw2ik8HPlelaVulg8iepVDsNRvWMYlz/uDrtePTSHszblMOazAI+X7uPqwfEn/I1tRYf/OxKdb9ucHusFjPBvjbG9I5h1tp9fLQykz6q8t7iNJLeTI41H72Wv91KoLdXSzZJRERERM5i4/rHEehtZfygdgT5NtPfoSYT+ISAdyDY/cHmC17epxWg13pkdHc6R7qyUM/pGMqL1yS7qrmbTBAYCwFRrmXj7AGupdr6XQ/3robzHgSLHTJ+ZETaQ7xve5F/GM/jP3M8/PdKeCUZlr5CfmEx1779M1tzSgj3t/H8lb3rrRUeFejN/43oAsAL326hqI0XV9uSXcSq3YewmE1ce1ThvetqUt5VQM4zNIzbTGrno8ceI0gXEREREWlJCWF+rH96lKebccq8vSx8eNsQ5m3KYWzfWOxWy4kPsgfAiCeh/0RY/DfI20xxeRW78ooxY9A1yImtOBPmP0XJgn/hX3Ejwb79eeuGAYQdY87+zcM68PGqPezMK2XGykzuOL9T/Z1K8iAnjcp966ku2I+vV+3YaE2KfHAC9P09+ASf0veiqXz4s2uptYuToog8qqD1oMQQOkX4sSOvlC/X7WfCkARPNfGspCC9mZxoJF1ERERERE5OVKA3159zCgFjSAJc8QYAAcCbH67mm7RszvEL5q/JGwn56a8kGvv40DaFkg6X4l9wEApqRulNZtf67xVFUHYIW9kh/um/h50FWcSstENWMFCzb0Ux5GyEkhwAbDWPBi18DvpPxBjyB6auqiAq0H7sYLi8CHb/BHtWQHUlmM1HphRY7RDRHeL6Q1C7RmctlFZUM2vtPoB631OTycR1g9vz1zmb+Whl5lkdpBuGUS+jorkpSG8mDa2RLiIicrZ64403ePHFF8nOzqZPnz68/vrrDB48uMF9p0+fzs0331znNbvdTnl5eUs0VUTOAo9d2oMftuTyc0YBF++Ow994iedCvua35V/jv+Mb2PHNcY/vDnS3ACXA5ob2MOEM7ch3eWFkGlGc3y2KpJhAVwBtOGHrXMjdBD+/ASvepFP1EBY6+3KxkUxEoJ9rTj0G7F8LOxfBvjWuteFPxDec0vBkNjra0Scxom62gXcwdLsEQjsCMDt1HyUV1XQM92Nop7B6p7qqfzx/+y6dDfuKSNtbSO/4oBO//xnE6TR4a8lONu4v5PXr+rVooK4gvZloJF1ERMTl448/ZvLkybz55psMGTKEV155hVGjRpGenk5kZGSDxwQGBpKefmQt4pYexRCRM1t8iC93XdCZqfO34jRgWK9O/Gb8+5gObIafXoHDB1yj5xhHqsbbA13p6T4hlJoDeGHhfgxMPPnbHtgsNensVjtE9IDIHqzLqeLufy4DIMO3HS+MTD7SgBFPwfYFsOw1TLsWM9ayjLGWZfDdsducbY3lh/JuFJsCGD8glmBvi6ttlcWQneYawT+cj1/mDwwG2NfASeY+CtG9cfQYy4Jf4oFAfj+kfd3fsZWH4cB2QvK38mr0Eipzt2H5+A2ICahbRd87GPyjwD+Sg+ZgvtxWxcheccQfHf9YvCCsC/hHNkldglOybw2sfh+qynFPNwDX5zn4DojsXu+QwsNVPDgzlfmbcwG4akA8F3ZruL9qDq0iSD+Zu+tHmzFjBtdddx1jx45l9uzZzd/Qk3CsNdJFRETONi+//DK33367e3T8zTffZM6cObz33ns88sgjDR5jMpmIjm6G6tMiIjX+cEFH8ksqiA324Q/nd3QVoYvuBVe9c8Jj/YC5K+aTW1zBuLih9G8fUm+fDft3u5+n7imou9Fkgi4joctIXvr3J7Tb9iFxpjxsZoP+7QKwGg7XyHlYZ+g4nM0+/Rg9PcN9+JL8MD64dUjd4LqqnOmzvmLHuqV0NGVhxsnFSdHEBNXMNT+wHXYtgew0LNlpvAcc8A4i5GcbLDdc72c4XUvX1RgNYAGKax7HEArcBLDrGDv4hEJUT4hMco3kmy1HBe0mV+2AiG4Q3hW8fKhyODGbTFjMpxHYOx3w06uuaQXO6oZ3WT0dZ78bsV70mOtGApC2t5C7PlzN3kNl2Kxmnrm8J8O7Rpx6O06Bx4P0U7m7DpCRkcFDDz3Eeeed14KtbZyySgcHSisBiA/29XBrREREPKeyspLVq1fz6KOPul8zm82MHDmS5cuXH/O4kpISEhIScDqd9O/fn+eff56ePXsec/+KigoqKircXxcVFTXNBYjIGcvby8Jfruh1yscnxQaSm57Hpv1FDQbpG/cdCXa35hRTWlGNXwPLL3+dG0FG9R3YrWYqKpw83rUHt5/fsc4+r/53NQDndg5nZcZBftp+gK/XZ3FZnyPr3W/IreDZtb44jYvp3z6YNZkFTN/vx7fXnHdkSejSA+xf8RnbFn3AUNMGwkyFUNrAxfmEQnhXjPAuzN7jy89ZTsICvLl/ZDdstSn0ZYegJIeNW7dzIDuTMFMRZpzEBvsQVLuKVVUpHNoNZQch40fX47hMGCGJ/FIcQb4plDHJcVjMRy3lF9oRuvwGQjsAsHr3QYJ8vOgcGVD3NIX7YNYf3O/n7PZb0u29WLengG15pRgGDDFvZpRlFeY171Oe+jH5fe/mp/DxPDFnB5UOJ+1CfZg2YQC94lo+zd/jQfqp3F13OBxMmDCBZ555hh9//JGCgoJjnt8TnXZtqru/3Uqgj8e/xSIiIh6Tn5+Pw+EgKiqqzutRUVFs2bKlwWO6devGe++9R3JyMoWFhbz00ksMHTqUjRs3Eh/f8JrEU6ZM4Zlnnmny9ouIHEtSTCCL0vPYlNVwfLFh/5Eg3WnAhn2FDOlYd+73odJKMg4cBuDBi7vy/DdbmL4sg5uHJWKtSaHfnlvC3E3ZADx5WRLfpGXxyvxt/OXrTQzvFkGAtxfVDicPf7YepwG/TY7h+XG9Gfn3xezKL+WNhdt58OJuAFTYg7llfXe2VD7MFd18mXpJKCazte6a9j6h4Odqpwm48HAlL7yyhJyCCkr2JfDs2CM3NtZkHuKahctxOA3O6RjKzzsP0q7ah/l3XHBkPnzlYchPJ2PzKuYtXEiMKZ8uEb50i/KvmVIAHD4IeZuh7BCmQ7sYWjskv7qBb+y3QFgX9kecx9T1Uew3wkgM8+WCrhFc0DWSxKrtGHMexFReQJXFhxnh9zJ12yAOHj6ylNyAhBBM7YL5Yv0C/lD+Hn3YSfyal7jCeJUxVgteNrBVgum9mvT++zdAYMwJ/080FY9GkKd6d/3ZZ58lMjKSW2+9lR9/PP7dGE902kfPR9ccOhERkZOTkpJCSkqK++uhQ4fSo0cP3nrrLf7yl780eMyjjz7K5MmT3V8XFRXRrl27BvcVEWkKSbGBAGzaXz9Ir6x2kp7tyg/vGRvIxv1FpO4pqBek16bBd4zw48aURN5avJN9BWXM3ZjDmGRXUPjm4h0YBvwmKYquUQG0D/Vl1tp97D5wmFfmb+OJ3ybx7tJdbNxfRJCPF09d1pNAby+eubwnd324hjcX7+DyPrF0iQrgtQXb2JJdTKifjT9fMxTTMZaZO1qwr42XrunDDe+u5D/Ld3NR90iGd4uktKKayR+n4nAajO0by5RxvRn+4iL2HCzjfysyuXmYa7Qbmy9l4cncsraQndWukf/IYjvL7rnIfSMCcAXspfn87b+zKdmbRgglDEoM4dxOoa5A2Vnlml+euRwObCP2wDY+qC2dXwKsqXngurmwztmR+yruIWN3DFBFZICdqwbEc/WAeDpF+APgvLQHP+8czwcL/8NFe6cRa8rHTpVr6nqdDHmDluTRIP1U7q4vXbqUd999l9TU1Ea9hyc6bc1HFxERcQkPD8disZCTk1Pn9ZycnEbPOffy8qJfv35s3779mPvY7Xbs9hP/sSki0lR6xLiC9C3ZRTicRp3501tziqlyGAT5eDEmOYaN+4tYt7eg3jnWZh4CoF+7ELy9LEw4J4HXFmzj3aU7GZMcw76CMmbXLJN293DXeuzeXhaeubwnN73/C9OXZTAoMZSX520F4M9jehAR4PpdeEmvaEb2iGT+5lwe/TyNRy/twbRFOwB4/spehDciQK91XpcIbhqayPRlGfzp0/XMvf98/jY3nYwDh4kJ8ubZsb3wtVm5f2RXHpuVxus/bOfqAfEE1KS9/23uFnbmlxIVaKfKYZBbXMGi9DxGJh0VB5pM5DgDeDMzFqfhCuZ7lQfy9UW/mt5cXoixYyFzPv8PfavXEWlzTTOucjipdhhUYeETx3DesV5LctcIrk4IYUBCKIMSQ+reFADMZhNDO0cytPNDHC67hwN5GYT52+tmFpjMrgJ5LahN5WIXFxdzww038PbbbxMeHt6oYzzRae8rcKWsxAZ7t+j7ioiItDY2m40BAwawYMECrrjiCgCcTicLFixg0qRJjTqHw+EgLS2NSy+9tBlbKiJychLD/PDxslBW5SDjQKl7dBZgY02qe6+4QPq2CwYgNbOg3jnW1oyk923v2ueGcxJ4c9EO1mQWsDbzEF+k7qfaaTC0Uxj9jpr3PrxbJJf0jOa7jdnc+YErJ3xY5zCuHnBkSpDJZOKZsb1YtmMxq3YfYuJ7K3EacEXfWC7pdfKp2w9f0p0ft+WxI6+UCe+scKf5//2aPgT5uILx3w2M552lO9mZV8pbi3fy0KhuLN9xgPd/ygDg/12VzNJt+byzdBcfr9pTN0gHPl+zD6cBnSP92Z5bwoZ9ReSXVNS9oeAdxI6IEUwq9cJmMZP6+G/wtVmx4arKvj7zICNDfLkzwt9VDLCRfH188G3f46S/L83BfOJdms/J3l3fsWMHGRkZXHbZZVitVqxWK//5z3/48ssvsVqt7Nixo6Waflz7C1zruMapaJyIiAiTJ0/m7bff5t///jebN2/mrrvuorS01F2P5sYbb6wz9e3ZZ5/l+++/Z+fOnaxZs4brr7+e3bt3c9ttt3nqEkRE6rGYTXSPcRUs+3XK+4Z9rq97xgaRHB+MyQT7C8vJLSp37+N0Gu509341gXxEgJ3L+7pGkV+et5UZv2QCcPfwzvXe/8nLkvC1ueZ9261mnr+yd72ptnHBPu756CUV1UQF2nnm8lMrludjszB1fF+sZpM7QL/t3A4M7Xxk8NRqMfOnUa4lzVzBegl//HQdANcNbsfwbpGMH+TKav5hSy65xUe+H4ZhMHP1Hvd5k2oyFX7anl+vLYvS8wAY0jEUX9uRcecgXy8u6u6aFnAyAXpr49Eg/ei767Vq764fPRetVvfu3UlLSyM1NdX9uPzyy7nwwgtJTU1tNXPPlO4uIiJyxPjx43nppZd48skn6du3L6mpqXz33Xfu6W6ZmZlkZWW59z906BC33347PXr04NJLL6WoqIhly5aRlJTkqUsQEWlQbSD56+JxtUXjesYG4m+30iXSNcp+9FJsO/NLKC6vxtvLTPfoI9XJb6mZy/3jtnzKq5wkxwcxrHPduewAscE+PDraFRA/dmkPEsL8GmzjxJQE+rUPxmI28f+uSibI1+sUrxaS44O5b0QXALpFBfDQqG719hnVM4oBCSGUVzm58p/L2HuojLhgHx4f4/od3iUqgH7tg3E4DT5fc2Qx9zWZBezMK8Xby8yY5BjO6+oK/hdvzav3HgvTXeuXD2/BtctbksfT3SdPnszEiRMZOHAggwcP5pVXXql3dz0uLo4pU6bg7e1Nr1517/wEBwcD1Hvdk44uHCciIiIwadKkY6a3L1q0qM7XU6dOZerUqS3QKhGR09NQ8bhqh5PNNUF77fJdfdsFszWnhHV7C7i4pytjeE1N+ntyfHCdudJJsYEM7RTGsh0HANdc9GMVo74hJZGrB7TDp2ZEvSFWi5kPbxvCwdJK4kNOP9N30kWdSYoNpE+74CNLux3FZDLxyOjuXPPmcgrLXBXVX7wmGf+jlp8bP7AdazML+OSXPfzh/I6YTCY+rRlFv7RXDAHeXlzQJYK3Fu/kx235GIbh/h6UVFSzctdBAC7s1rLrl7cUj46kw8nfXW/tqh1OsmvSWOI1ki4iIiIicsbq0cBI+s78UsqrnPjZLHSoGd3uUzsv/aiR9LU1QXptqvvR7qhZJ71rlD8XJx2/yObxAvRavjZrkwTo4ArCR/SIOm7huUGJoVxcM9/8pqGJDO1Ut57Yb/vE4muzsDO/lFW7D1FW6eCrda6Y75qBruzoAYkh+HhZyCuuYEtNpXyAZdvzqXIYJIT50iG84eyBts7jI+lwcnfXf2369OlN36DTkF1UjsNpYLOYiTiJiokiIiIiItK2dI8OwGSCvOIK8ooriAiws2GfK9U9KTbQPS+6tnjc+j2FOJ0GZrPpyHz0mqJxRxveLZJP/pBCYphvm51b/fL4vizfcaDB0W5/u5UxvWOYuXovH/+yh72dD1NSUU27UB+GdAgFwG61cE7HUBam57Fka577hsjCmvnoF3aLPGOXu/b4SPqZpnY+ekywd5v9gRIRERERkRPztVndo7m1Ke5HF42r1S0qAG8vM8UV1ezML6G0opr0bNd+R1dtP9rgDqFEBrbd1aL87VZ+kxRVb9mzWrUF5Oasz+Lfy3YDcHX/dnViqPO6uAL8H7e5iscZhsEi93z0MzPVHRSkNznNRxcREREROXv8unjcBvfya0eCdKvFTO+ar1P3FLJ+byFOA2KDvIlqw4H46RiQEELHCD/Kqhyk7inAZIKrBsTV2ef8rq5AfGXGQcoqHaTnFJNVWI7dauacjvWL6Z0pFKQ3sdqR9FgF6SIiIiIiZzz3vPT9RTidhruIXK+4wDr79YkPBmDdngLW7jkEHFkf/WxkMpkYP/DI6lxDO4XVmzffKcKP2CBvKqudrNh1wL302tBOYQ0WrTtTKEhvYhpJFxERERE5e7grvGcVsfuga2613Wqmc4R/nf1qA/LUPQWkuovGNZzqfrYY1z8ea016+zUD6i+nbTKZ3KPpS7bms3CLK9X9wu5n5tJrtVpF4bgzRU5ROXPSXFUJu0YFnGBvERERERFp63rWjKTvzCthVYZrabDuMYH15mLXjqRvzipyD+w1VDTubBIRYOfPY3qwJbuY0b0brmJ/ftcIZvyyh+83ZZNd6FpFa3hXBenSCIZh8PisNIrLq+kTH8SonlGebpKIiIiIiDSziAA74f428ksq+XzNPgB6xQbW2y8+xMe938HSSqxmU51562erm4Z1OO72YZ3CMZtgb8204k4RfrQPa5rl5Forpbs3kS/X7Wf+5ly8LCb+dnWfY1YxFBERERGRM4fJZHLPS1++8wBAg8G3yWRyL8UGrrnsZ/K86qYS5OvlXmceXMvTnekUSTaBAyUVPPPVJgDuubAz3aKV6i4iIiIicraorfBeq1dswyPktSnvoFT3k1G7FBu41kc/0ylIbwJPf7WJg6WVdI8O4O7hnT3dHBERERERaUFJR6W3W80mukb7N7jf0dXcFaQ3Xu2a6H42C4M6nPnF9jQn/TTN25TDV+v2YzbB365OxmbVfQ8RERERkbPJ0SPpXaMCsFsbTmNPjg/GZALDUGX3k9GvXTB/vaIX7UN9j/m9PZMoSD8NhWVVPD4rDYDbz+9I8lHpKyIiIiIicnboEO6H3WqmotpZb330owX5ePH/xiVTUlFNYrhfC7awbTOZTFx/ToKnm9FiFKSfhufnbCa3uIIO4X48MLKrp5sjIiIiIiIeYLWY6R4dwLq9hSes2P67QfXXAxc5mnKzT8OF3SOIDLDz/65KVmVGEREREZGz2J8u6c7VA+IZ1z/e002RNk4j6afhkl4xDO8WqQBdREREROQsN6xzOMM6h3u6GXIG0Ej6aVKALiIiIiIiIk1FQbqIiIiIiIhIK6EgXURERERERKSVUJAuIiIiIiIi0kooSBcRERERERFpJRSki4iIiIiIiLQSCtJFREREREREWgkF6SIiIiIiIiKthIJ0ERERERERkVZCQbqIiIiIiIhIK6EgXURERERERKSVsHq6AS3NMAwAioqKPNwSERERl9o+qbaPktOn/l5ERFqTk+nrz7ogvbi4GIB27dp5uCUiIiJ1FRcXExQU5OlmnBHU34uISGvUmL7eZJxlt+2dTif79+8nICAAk8l0WucqKiqiXbt27Nmzh8DAwCZqYcvSNXheW28/6Bpag7befji7r8EwDIqLi4mNjcVs1ky0pqD+/oi23n7QNbQGbb39oGtoDdp6+6Fl+vqzbiTdbDYTHx/fpOcMDAxss//JaukaPK+ttx90Da1BW28/nL3XoBH0pqX+vr623n7QNbQGbb39oGtoDdp6+6F5+3rdrhcRERERERFpJRSki4iIiIiIiLQSCtJPg91u56mnnsJut3u6KadM1+B5bb39oGtoDdp6+0HXIK1XW/9c23r7QdfQGrT19oOuoTVo6+2HlrmGs65wnIiIiIiIiEhrpZF0ERERERERkVZCQbqIiIiIiIhIK6EgXURERERERKSVUJAuIiIiIiIi0kooSD8Nb7zxBomJiXh7ezNkyBBWrlzp6SYd05IlS7jsssuIjY3FZDIxe/bsOtsNw+DJJ58kJiYGHx8fRo4cybZt2zzT2AZMmTKFQYMGERAQQGRkJFdccQXp6el19ikvL+eee+4hLCwMf39/rrrqKnJycjzU4vqmTZtGcnIygYGBBAYGkpKSwrfffuve3trb/2svvPACJpOJ+++/3/1aa7+Gp59+GpPJVOfRvXt39/bW3v5a+/bt4/rrrycsLAwfHx969+7NqlWr3Ntb889zYmJivc/AZDJxzz33AG3jM3A4HDzxxBN06NABHx8fOnXqxF/+8heOrsPamj8DOTnq61uO+vrWR32957Tlvh7afn/v8b7ekFMyY8YMw2azGe+9956xceNG4/bbbzeCg4ONnJwcTzetQd98843x+OOPG59//rkBGLNmzaqz/YUXXjCCgoKM2bNnG+vWrTMuv/xyo0OHDkZZWZlnGvwro0aNMt5//31jw4YNRmpqqnHppZca7du3N0pKStz73HnnnUa7du2MBQsWGKtWrTLOOeccY+jQoR5sdV1ffvmlMWfOHGPr1q1Genq68dhjjxleXl7Ghg0bDMNo/e0/2sqVK43ExEQjOTnZuO+++9yvt/ZreOqpp4yePXsaWVlZ7kdeXp57e2tvv2EYxsGDB42EhATjpptuMlasWGHs3LnTmDt3rrF9+3b3Pq355zk3N7fO93/evHkGYCxcuNAwjLbxGTz33HNGWFiY8fXXXxu7du0yZs6cafj7+xuvvvqqe5/W/BlI46mvb1nq61sX9fWe09b7esNo+/29p/t6BemnaPDgwcY999zj/trhcBixsbHGlClTPNiqxvl1x+10Oo3o6GjjxRdfdL9WUFBg2O1246OPPvJAC08sNzfXAIzFixcbhuFqr5eXlzFz5kz3Pps3bzYAY/ny5Z5q5gmFhIQY77zzTptqf3FxsdGlSxdj3rx5xgUXXODuuNvCNTz11FNGnz59GtzWFtpvGIbx8MMPG+eee+4xt7e1n+f77rvP6NSpk+F0OtvMZzBmzBjjlltuqfPauHHjjAkTJhiG0fY+Azk29fWepb7ec9TXe9aZ1tcbRtvr7z3d1yvd/RRUVlayevVqRo4c6X7NbDYzcuRIli9f7sGWnZpdu3aRnZ1d53qCgoIYMmRIq72ewsJCAEJDQwFYvXo1VVVVda6he/futG/fvlVeg8PhYMaMGZSWlpKSktKm2n/PPfcwZsyYOm2FtvMZbNu2jdjYWDp27MiECRPIzMwE2k77v/zySwYOHMg111xDZGQk/fr14+2333Zvb0s/z5WVlXzwwQfccsstmEymNvMZDB06lAULFrB161YA1q1bx9KlSxk9ejTQtj4DOTb19Z6nvt5z1Nd71pnU10Pb7O893ddbT/sMZ6H8/HwcDgdRUVF1Xo+KimLLli0eatWpy87OBmjwemq3tSZOp5P777+fYcOG0atXL8B1DTabjeDg4Dr7trZrSEtLIyUlhfLycvz9/Zk1axZJSUmkpqa2ifbPmDGDNWvW8Msvv9Tb1hY+gyFDhjB9+nS6detGVlYWzzzzDOeddx4bNmxoE+0H2LlzJ9OmTWPy5Mk89thj/PLLL/zf//0fNpuNiRMntqmf59mzZ1NQUMBNN90EtI3/QwCPPPIIRUVFdO/eHYvFgsPh4LnnnmPChAlA2/udKg1TX+9Z6us9R329551JfT20zf7e0329gnRpc+655x42bNjA0qVLPd2Uk9atWzdSU1MpLCzk008/ZeLEiSxevNjTzWqUPXv2cN999zFv3jy8vb093ZxTUnv3EyA5OZkhQ4aQkJDAJ598go+Pjwdb1nhOp5OBAwfy/PPPA9CvXz82bNjAm2++ycSJEz3cupPz7rvvMnr0aGJjYz3dlJPyySef8OGHH/K///2Pnj17kpqayv33309sbGyb+wxEWiv19Z6hvr51OJP6emib/b2n+3qlu5+C8PBwLBZLvQqEOTk5REdHe6hVp662zW3heiZNmsTXX3/NwoULiY+Pd78eHR1NZWUlBQUFdfZvbddgs9no3LkzAwYMYMqUKfTp04dXX321TbR/9erV5Obm0r9/f6xWK1arlcWLF/Paa69htVqJiopq9dfwa8HBwXTt2pXt27e3ic8AICYmhqSkpDqv9ejRw53K11Z+nnfv3s38+fO57bbb3K+1lc/gj3/8I4888gjXXnstvXv35oYbbuCBBx5gypQpQNv5DOT41Nd7jvp6z1Ff3zqcKX09tN3+3tN9vYL0U2Cz2RgwYAALFixwv+Z0OlmwYAEpKSkebNmp6dChA9HR0XWup6ioiBUrVrSa6zEMg0mTJjFr1ix++OEHOnToUGf7gAED8PLyqnMN6enpZGZmtppraIjT6aSioqJNtH/EiBGkpaWRmprqfgwcOJAJEya4n7f2a/i1kpISduzYQUxMTJv4DACGDRtWb0mirVu3kpCQALSNn2eA999/n8jISMaMGeN+ra18BocPH8Zsrtt9WiwWnE4n0HY+Azk+9fUtT32956mvbx3OlL4e2m5/7/G+/rRLz52lZsyYYdjtdmP69OnGpk2bjDvuuMMIDg42srOzPd20BhUXFxtr16411q5dawDGyy+/bKxdu9bYvXu3YRiuJQSCg4ONL774wli/fr0xduzYVrWMw1133WUEBQUZixYtqrOcw+HDh9373HnnnUb79u2NH374wVi1apWRkpJipKSkeLDVdT3yyCPG4sWLjV27dhnr1683HnnkEcNkMhnff/+9YRitv/0NObriq2G0/mt48MEHjUWLFhm7du0yfvrpJ2PkyJFGeHi4kZubaxhG62+/YbiWxLFarcZzzz1nbNu2zfjwww8NX19f44MPPnDv09p/nh0Oh9G+fXvj4YcfrretLXwGEydONOLi4tzLsnz++edGeHi48ac//cm9T2v/DKRx1Ne3LPX1rZP6+pZ3JvT1htG2+3tP9/UK0k/D66+/brRv396w2WzG4MGDjZ9//tnTTTqmhQsXGkC9x8SJEw3DcC0j8MQTTxhRUVGG3W43RowYYaSnp3u20UdpqO2A8f7777v3KSsrM+6++24jJCTE8PX1Na688kojKyvLc43+lVtuucVISEgwbDabERERYYwYMcLdaRtG629/Q37dcbf2axg/frwRExNj2Gw2Iy4uzhg/fnydNUdbe/trffXVV0avXr0Mu91udO/e3fjXv/5VZ3tr/3meO3euATTYprbwGRQVFRn33Xef0b59e8Pb29vo2LGj8fjjjxsVFRXufVr7ZyCNp76+5aivb53U13tGW+/rDaNt9/ee7utNhmEYpz8eLyIiIiIiIiKnS3PSRURERERERFoJBekiIiIiIiIirYSCdBEREREREZFWQkG6iIiIiIiISCuhIF1ERERERESklVCQLiIiIiIiItJKKEgXERERERERaSUUpIuIiIiIiIi0EgrSRaTFmUwmZs+e7elmiIiISDNRXy9y6hSki5xlbrrpJkwmU73HJZdc4ummiYiISBNQXy/Stlk93QARaXmXXHIJ77//fp3X7Ha7h1ojIiIiTU19vUjbpZF0kbOQ3W4nOjq6ziMkJARwpadNmzaN0aNH4+PjQ8eOHfn000/rHJ+WlsZFF12Ej48PYWFh3HHHHZSUlNTZ57333qNnz57Y7XZiYmKYNGlSne35+flceeWV+Pr60qVLF7788svmvWgREZGziPp6kbZLQbqI1PPEE09w1VVXsW7dOiZMmMC1117L5s2bASgtLWXUqFGEhITwyy+/MHPmTObPn1+nY542bRr33HMPd9xxB2lpaXz55Zd07ty5zns888wz/O53v2P9+vVceumlTJgwgYMHD7bodYqIiJyt1NeLtGKGiJxVJk6caFgsFsPPz6/O47nnnjMMwzAA484776xzzJAhQ4y77rrLMAzD+Ne//mWEhIQYJSUl7u1z5swxzGazkZ2dbRiGYcTGxhqPP/74MdsAGH/+85/dX5eUlBiA8e233zbZdYqIiJyt1NeLtG2aky5yFrrwwguZNm1anddCQ0Pdz1NSUupsS0lJITU1FYDNmzfTp08f/Pz83NuHDRuG0+kkPT0dk8nE/v37GTFixHHbkJyc7H7u5+dHYGAgubm5p3pJIiIichT19SJtl4J0kbOQn59fvZS0puLj49Oo/by8vOp8bTKZcDqdzdEkERGRs476epG2S3PSRaSen3/+ud7XPXr0AKBHjx6sW7eO0tJS9/affvoJs9lMt27dCAgIIDExkQULFrRom0VERKTx1NeLtF4aSRc5C1VUVJCdnV3nNavVSnh4OAAzZ85k4MCBnHvuuXz44YesXLmSd999F4AJEybw1FNPMXHiRJ5++mny8vK49957ueGGG4iKigLg6aef5s477yQyMpLRo0dTXFzMTz/9xL333tuyFyoiInKWUl8v0nYpSBc5C3333XfExMTUea1bt25s2bIFcFVjnTFjBnfffTcxMTF89NFHJCUlAeDr68vcuXO57777GDRoEL6+vlx11VW8/PLL7nNNnDiR8vJypk6dykMPPUR4eDhXX311y12giIjIWU59vUjbZTIMw/B0I0Sk9TCZTMyaNYsrrrjC000RERGRZqC+XqR105x0ERERERERkVZCQbqIiIiIiIhIK6F0dxEREREREZFWQiPpIiIiIiIiIq2EgnQRERERERGRVkJBuoiIiIiIiEgroSBdREREREREpJVQkC4iIiIiIiLSSihIFxEREREREWklFKSLiIiIiIiItBIK0kVERERERERaif8PcsTVdmyaqI0AAAAASUVORK5CYII=",
      "text/plain": [
       "<Figure size 1200x400 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# cell1: Importing necessary libraries, loading and preprocessing the CIFAR-10 dataset, defining and training a CNN model with data augmentation and learning rate scheduling\n",
    "\n",
    "import numpy as np  # Importing numpy for numerical operations\n",
    "import matplotlib.pyplot as plt  # Importing matplotlib for plotting\n",
    "from keras.models import Sequential  # Importing Sequential model from Keras for building a linear stack of layers\n",
    "from keras.layers import Conv2D, MaxPooling2D, Flatten, Dense, Dropout, BatchNormalization  # Importing layers for building the CNN\n",
    "from keras.callbacks import LearningRateScheduler, EarlyStopping, ReduceLROnPlateau  # Importing callbacks for learning rate scheduling and early stopping\n",
    "from keras.datasets import cifar10  # Importing the CIFAR-10 dataset\n",
    "from keras.utils import to_categorical  # Importing utility for converting labels to one-hot encoding\n",
    "from tensorflow.keras.preprocessing.image import ImageDataGenerator  # Importing ImageDataGenerator for data augmentation\n",
    "from keras.optimizers import Adam  # Importing Adam optimizer\n",
    "from keras.regularizers import l2  # Importing L2 regularization\n",
    "\n",
    "# Load and preprocess the CIFAR-10 dataset\n",
    "(x_train, y_train), (x_val, y_val) = cifar10.load_data()  # Loading the CIFAR-10 dataset\n",
    "\n",
    "x_train = x_train.astype('float32') / 255.0  # Converting pixel values to float32 and normalizing to [0, 1]\n",
    "x_val = x_val.astype('float32') / 255.0  # Converting pixel values to float32 and normalizing to [0, 1]\n",
    "y_train = to_categorical(y_train, 10)  # Converting labels to one-hot encoding\n",
    "y_val = to_categorical(y_val, 10)  # Converting labels to one-hot encoding\n",
    "\n",
    "# Enhanced Data Augmentation\n",
    "datagen = ImageDataGenerator(  # Creating an ImageDataGenerator for data augmentation\n",
    "    width_shift_range=0.2,  # Allowing horizontal shifts\n",
    "    height_shift_range=0.2,  # Allowing vertical shifts\n",
    "    horizontal_flip=True,  # Allowing horizontal flips\n",
    "    rotation_range=20,  # Allowing rotations up to 20 degrees\n",
    "    zoom_range=0.2,  # Allowing zooming\n",
    "    shear_range=0.2  # Allowing shearing\n",
    ")\n",
    "datagen.fit(x_train)  # Fitting the data generator on the training data\n",
    "\n",
    "# Slightly More Complex Model\n",
    "model = Sequential()  # Creating a Sequential model\n",
    "model.add(Conv2D(64, (3, 3), padding='same', activation='relu', input_shape=(32, 32, 3), name='conv2d_1'))  # Adding a convolutional layer with 64 filters\n",
    "model.add(BatchNormalization(name='batch_norm_1'))  # Adding batch normalization to stabilize and speed up training\n",
    "model.add(Conv2D(64, (3, 3), padding='same', activation='relu', name='conv2d_2'))  # Adding another convolutional layer with 64 filters\n",
    "model.add(BatchNormalization(name='batch_norm_2'))  # Adding batch normalization\n",
    "model.add(MaxPooling2D(pool_size=(2, 2), name='max_pool_1'))  # Adding a max pooling layer to reduce spatial dimensions\n",
    "\n",
    "model.add(Conv2D(128, (3, 3), padding='same', activation='relu', name='conv2d_3'))  # Adding a convolutional layer with 128 filters\n",
    "model.add(BatchNormalization(name='batch_norm_3'))  # Adding batch normalization\n",
    "model.add(Conv2D(128, (3, 3), padding='same', activation='relu', name='conv2d_4'))  # Adding another convolutional layer with 128 filters\n",
    "model.add(BatchNormalization(name='batch_norm_4'))  # Adding batch normalization\n",
    "model.add(MaxPooling2D(pool_size=(2, 2), name='max_pool_2'))  # Adding a max pooling layer\n",
    "\n",
    "model.add(Conv2D(256, (3, 3), padding='same', activation='relu', name='conv2d_5'))  # Adding a convolutional layer with 256 filters\n",
    "model.add(BatchNormalization(name='batch_norm_5'))  # Adding batch normalization\n",
    "model.add(Conv2D(256, (3, 3), padding='same', activation='relu', name='conv2d_6'))  # Adding another convolutional layer with 256 filters\n",
    "model.add(BatchNormalization(name='batch_norm_6'))  # Adding batch normalization\n",
    "model.add(MaxPooling2D(pool_size=(2, 2), name='max_pool_3'))  # Adding a max pooling layer\n",
    "\n",
    "model.add(Flatten(name='flatten'))  # Flattening the output to feed into dense layers\n",
    "model.add(Dense(512, activation='relu', kernel_regularizer=l2(0.001), name='dense_1'))  # Adding a dense layer with 512 units and L2 regularization\n",
    "model.add(Dropout(0.5, name='dropout_1'))  # Adding dropout for regularization\n",
    "model.add(Dense(10, activation='softmax', name='dense_2'))  # Adding the output layer with softmax activation\n",
    "\n",
    "# Compile the model with a reduced learning rate\n",
    "model.compile(optimizer=Adam(learning_rate=1e-4), loss='categorical_crossentropy', metrics=['accuracy'])  # Compiling the model with Adam optimizer and categorical cross-entropy loss\n",
    "model.summary()  # Printing a summary of the model architecture\n",
    "\n",
    "# Adjusted Learning Rate Scheduler\n",
    "def lr_schedule(epoch):  # Defining a learning rate schedule function\n",
    "    lr = 1e-4  # Starting learning rate\n",
    "    if epoch > 50:  # If epoch number is greater than 50\n",
    "        lr *= 0.1  # Reduce learning rate by a factor of 0.1\n",
    "    elif epoch > 30:  # If epoch number is greater than 30 but less than or equal to 50\n",
    "        lr *= 0.5  # Reduce learning rate by a factor of 0.5\n",
    "    print(f\"Learning rate for epoch {epoch + 1} is {lr}\")  # Printing the learning rate for the current epoch\n",
    "    return lr  # Returning the adjusted learning rate\n",
    "\n",
    "lr_scheduler = LearningRateScheduler(lr_schedule)  # Creating a LearningRateScheduler callback\n",
    "early_stopper = EarlyStopping(monitor='val_loss', patience=10, restore_best_weights=True)  # Creating an EarlyStopping callback\n",
    "reduce_lr = ReduceLROnPlateau(monitor='val_loss', factor=0.5, patience=5, min_lr=1e-7, verbose=1)  # Creating a ReduceLROnPlateau callback\n",
    "\n",
    "# Training the model\n",
    "history = model.fit(  # Training the model\n",
    "    datagen.flow(x_train, y_train, batch_size=64),  # Using the data generator to generate augmented data\n",
    "    epochs=80,  # Number of epochs to train\n",
    "    validation_data=(x_val, y_val),  # Validation data\n",
    "    callbacks=[lr_scheduler, early_stopper, reduce_lr],  # Using the defined callbacks\n",
    "    steps_per_epoch=x_train.shape[0] // 64,  # Number of steps per epoch\n",
    "    verbose=1  # Verbosity level\n",
    ")\n",
    "\n",
    "# Ensure the model is built by calling predict on a dummy input\n",
    "model.predict(np.expand_dims(x_val[0], axis=0))  # Predicting on a dummy input to build the model\n",
    "\n",
    "# Plotting Training and Validation Curves\n",
    "plt.figure(figsize=(12, 4))  # Creating a figure for plotting\n",
    "\n",
    "plt.subplot(1, 2, 1)  # Creating a subplot for accuracy\n",
    "plt.plot(history.history['accuracy'])  # Plotting training accuracy\n",
    "plt.plot(history.history['val_accuracy'])  # Plotting validation accuracy\n",
    "plt.title('Model Accuracy')  # Setting the title\n",
    "plt.xlabel('Epoch')  # Setting the x-axis label\n",
    "plt.ylabel('Accuracy')  # Setting the y-axis label\n",
    "plt.legend(['Train', 'Val'], loc='upper left')  # Adding a legend\n",
    "\n",
    "plt.subplot(1, 2, 2)  # Creating a subplot for loss\n",
    "plt.plot(history.history['loss'])  # Plotting training loss\n",
    "plt.plot(history.history['val_loss'])  # Plotting validation loss\n",
    "plt.title('Model Loss')  # Setting the title\n",
    "plt.xlabel('Epoch')  # Setting the x-axis label\n",
    "plt.ylabel('Loss')  # Setting the y-axis label\n",
    "plt.legend(['Train', 'Val'], loc='upper left')  # Adding a legend\n",
    "\n",
    "plt.show()  # Displaying the plots"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/matt/CIFAR10_CNN_Project/venv/lib/python3.12/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
      "  from .autonotebook import tqdm as notebook_tqdm\n",
      "[I 2024-09-02 17:24:50,570] A new study created in memory with name: no-name-d2663e73-aaa8-4770-9036-13b8799583b8\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/matt/CIFAR10_CNN_Project/venv/lib/python3.12/site-packages/keras/src/layers/convolutional/base_conv.py:107: UserWarning: Do not pass an `input_shape`/`input_dim` argument to a layer. When using Sequential models, prefer using an `Input(shape)` object as the first layer in the model instead.\n",
      "  super().__init__(activity_regularizer=activity_regularizer, **kwargs)\n",
      "/home/matt/CIFAR10_CNN_Project/venv/lib/python3.12/site-packages/keras/src/trainers/data_adapters/py_dataset_adapter.py:122: UserWarning: Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored.\n",
      "  self._warn_if_super_not_called()\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m1559/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 8ms/step - accuracy: 0.3068 - loss: 2.9694\n",
      "Epoch 1: val_loss improved from inf to 2.29696, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m23s\u001b[0m 11ms/step - accuracy: 0.3070 - loss: 2.9686 - val_accuracy: 0.4700 - val_loss: 2.2970 - learning_rate: 1.0000e-04\n",
      "Epoch 2/15\n",
      "\u001b[1m   1/1562\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m6s\u001b[0m 4ms/step - accuracy: 0.6250 - loss: 2.0804"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/matt/miniconda3/lib/python3.12/contextlib.py:158: UserWarning: Your input ran out of data; interrupting training. Make sure that your dataset or generator can generate at least `steps_per_epoch * epochs` batches. You may need to use the `.repeat()` function when building your dataset.\n",
      "  self.gen.throw(value)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Epoch 2: val_loss improved from 2.29696 to 2.29153, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 354us/step - accuracy: 0.6250 - loss: 2.0804 - val_accuracy: 0.4708 - val_loss: 2.2915 - learning_rate: 1.0000e-04\n",
      "Epoch 3/15\n",
      "\u001b[1m1555/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 6ms/step - accuracy: 0.4506 - loss: 2.2890\n",
      "Epoch 3: val_loss improved from 2.29153 to 1.81712, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m10s\u001b[0m 6ms/step - accuracy: 0.4507 - loss: 2.2886 - val_accuracy: 0.5703 - val_loss: 1.8171 - learning_rate: 1.0000e-04\n",
      "Epoch 4/15\n",
      "\u001b[1m   1/1562\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m5s\u001b[0m 4ms/step - accuracy: 0.5312 - loss: 2.1078\n",
      "Epoch 4: val_loss did not improve from 1.81712\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 227us/step - accuracy: 0.5312 - loss: 2.1078 - val_accuracy: 0.5680 - val_loss: 1.8196 - learning_rate: 1.0000e-04\n",
      "Epoch 5/15\n",
      "\u001b[1m1555/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 6ms/step - accuracy: 0.5107 - loss: 1.9611\n",
      "Epoch 5: val_loss improved from 1.81712 to 1.61855, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m10s\u001b[0m 6ms/step - accuracy: 0.5108 - loss: 1.9607 - val_accuracy: 0.5961 - val_loss: 1.6186 - learning_rate: 1.0000e-04\n",
      "Epoch 6/15\n",
      "\u001b[1m   1/1562\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m5s\u001b[0m 4ms/step - accuracy: 0.5000 - loss: 1.6931\n",
      "Epoch 6: val_loss did not improve from 1.61855\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 252us/step - accuracy: 0.5000 - loss: 1.6931 - val_accuracy: 0.5953 - val_loss: 1.6205 - learning_rate: 1.0000e-04\n",
      "Epoch 7/15\n",
      "\u001b[1m1560/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 6ms/step - accuracy: 0.5603 - loss: 1.6938\n",
      "Epoch 7: val_loss improved from 1.61855 to 1.44462, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m10s\u001b[0m 6ms/step - accuracy: 0.5603 - loss: 1.6937 - val_accuracy: 0.6206 - val_loss: 1.4446 - learning_rate: 1.0000e-04\n",
      "Epoch 8/15\n",
      "\u001b[1m   1/1562\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m6s\u001b[0m 4ms/step - accuracy: 0.5938 - loss: 1.5169\n",
      "Epoch 8: val_loss improved from 1.44462 to 1.44456, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 328us/step - accuracy: 0.5938 - loss: 1.5169 - val_accuracy: 0.6197 - val_loss: 1.4446 - learning_rate: 1.0000e-04\n",
      "Epoch 9/15\n",
      "\u001b[1m1554/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 6ms/step - accuracy: 0.5980 - loss: 1.4678\n",
      "Epoch 9: val_loss improved from 1.44456 to 1.19925, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 6ms/step - accuracy: 0.5981 - loss: 1.4676 - val_accuracy: 0.6786 - val_loss: 1.1992 - learning_rate: 1.0000e-04\n",
      "Epoch 10/15\n",
      "\u001b[1m   1/1562\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m5s\u001b[0m 4ms/step - accuracy: 0.7500 - loss: 1.2794"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:25:53.565747: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 7338431921674375828\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Epoch 10: val_loss did not improve from 1.19925\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 225us/step - accuracy: 0.7500 - loss: 1.2794 - val_accuracy: 0.6773 - val_loss: 1.2008 - learning_rate: 1.0000e-04\n",
      "Epoch 11/15\n",
      "\u001b[1m1553/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 6ms/step - accuracy: 0.6327 - loss: 1.3300\n",
      "Epoch 11: val_loss improved from 1.19925 to 1.14776, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m10s\u001b[0m 6ms/step - accuracy: 0.6328 - loss: 1.3298 - val_accuracy: 0.6871 - val_loss: 1.1478 - learning_rate: 1.0000e-04\n",
      "Epoch 12/15\n",
      "\u001b[1m   1/1562\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m5s\u001b[0m 3ms/step - accuracy: 0.7812 - loss: 0.9753\n",
      "Epoch 12: val_loss improved from 1.14776 to 1.14555, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 315us/step - accuracy: 0.7812 - loss: 0.9753 - val_accuracy: 0.6884 - val_loss: 1.1456 - learning_rate: 1.0000e-04\n",
      "Epoch 13/15\n",
      "\u001b[1m1555/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 6ms/step - accuracy: 0.6566 - loss: 1.2090\n",
      "Epoch 13: val_loss improved from 1.14555 to 1.06226, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m10s\u001b[0m 6ms/step - accuracy: 0.6566 - loss: 1.2089 - val_accuracy: 0.7155 - val_loss: 1.0623 - learning_rate: 1.0000e-04\n",
      "Epoch 14/15\n",
      "\u001b[1m   1/1562\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m6s\u001b[0m 4ms/step - accuracy: 0.7500 - loss: 1.1109\n",
      "Epoch 14: val_loss improved from 1.06226 to 1.06217, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 345us/step - accuracy: 0.7500 - loss: 1.1109 - val_accuracy: 0.7150 - val_loss: 1.0622 - learning_rate: 1.0000e-04\n",
      "Epoch 15/15\n",
      "\u001b[1m1553/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 6ms/step - accuracy: 0.6857 - loss: 1.1252\n",
      "Epoch 15: val_loss improved from 1.06217 to 0.88902, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m10s\u001b[0m 6ms/step - accuracy: 0.6857 - loss: 1.1251 - val_accuracy: 0.7578 - val_loss: 0.8890 - learning_rate: 1.0000e-04\n",
      "Restoring model weights from the end of the best epoch: 15.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-09-02 17:26:24,135] Trial 0 finished with value: 0.7577999830245972 and parameters: {'lr': 0.0008750803832793538, 'batch_size': 32}. Best is trial 0 with value: 0.7577999830245972.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:26:27.270972: I external/local_xla/xla/stream_executor/cuda/cuda_asm_compiler.cc:393] ptxas warning : Registers are spilled to local memory in function 'gemm_fusion_dot_1883', 20 bytes spill stores, 20 bytes spill loads\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m370/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━━\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.2883 - loss: 3.1595"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:26:38.814142: I external/local_xla/xla/stream_executor/cuda/cuda_asm_compiler.cc:393] ptxas warning : Registers are spilled to local memory in function 'gemm_fusion_dot_1883', 12 bytes spill stores, 12 bytes spill loads\n",
      "\n",
      "2024-09-02 17:26:38.975900: I external/local_xla/xla/stream_executor/cuda/cuda_asm_compiler.cc:393] ptxas warning : Registers are spilled to local memory in function 'gemm_fusion_dot_2871', 4 bytes spill stores, 4 bytes spill loads\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 30ms/step - accuracy: 0.2913 - loss: 3.1395\n",
      "Epoch 1: val_loss improved from inf to 3.55285, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m21s\u001b[0m 36ms/step - accuracy: 0.2914 - loss: 3.1385 - val_accuracy: 0.2334 - val_loss: 3.5528 - learning_rate: 1.0000e-04\n",
      "Epoch 2/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m3s\u001b[0m 9ms/step - accuracy: 0.3984 - loss: 2.4994\n",
      "Epoch 2: val_loss improved from 3.55285 to 3.54139, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 1ms/step - accuracy: 0.3984 - loss: 2.4994 - val_accuracy: 0.2374 - val_loss: 3.5414 - learning_rate: 1.0000e-04\n",
      "Epoch 3/15\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.4282 - loss: 2.4360\n",
      "Epoch 3: val_loss improved from 3.54139 to 2.10201, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.4282 - loss: 2.4359 - val_accuracy: 0.5319 - val_loss: 2.1020 - learning_rate: 1.0000e-04\n",
      "Epoch 4/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 6ms/step - accuracy: 0.5234 - loss: 2.1903\n",
      "Epoch 4: val_loss did not improve from 2.10201\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 910us/step - accuracy: 0.5234 - loss: 2.1903 - val_accuracy: 0.5299 - val_loss: 2.1082 - learning_rate: 1.0000e-04\n",
      "Epoch 5/15\n",
      "\u001b[1m385/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.4838 - loss: 2.2348\n",
      "Epoch 5: val_loss improved from 2.10201 to 1.92491, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.4839 - loss: 2.2344 - val_accuracy: 0.5715 - val_loss: 1.9249 - learning_rate: 1.0000e-04\n",
      "Epoch 6/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 7ms/step - accuracy: 0.5234 - loss: 2.1754\n",
      "Epoch 6: val_loss improved from 1.92491 to 1.92261, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 1ms/step - accuracy: 0.5234 - loss: 2.1754 - val_accuracy: 0.5722 - val_loss: 1.9226 - learning_rate: 1.0000e-04\n",
      "Epoch 7/15\n",
      "\u001b[1m385/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.5130 - loss: 2.0798\n",
      "Epoch 7: val_loss did not improve from 1.92261\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.5131 - loss: 2.0794 - val_accuracy: 0.5622 - val_loss: 1.9261 - learning_rate: 1.0000e-04\n",
      "Epoch 8/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 6ms/step - accuracy: 0.5547 - loss: 1.9210\n",
      "Epoch 8: val_loss did not improve from 1.92261\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 899us/step - accuracy: 0.5547 - loss: 1.9210 - val_accuracy: 0.5593 - val_loss: 1.9417 - learning_rate: 1.0000e-04\n",
      "Epoch 9/15\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.5463 - loss: 1.9394\n",
      "Epoch 9: val_loss improved from 1.92261 to 1.84829, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.5463 - loss: 1.9393 - val_accuracy: 0.5684 - val_loss: 1.8483 - learning_rate: 1.0000e-04\n",
      "Epoch 10/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 6ms/step - accuracy: 0.5625 - loss: 1.8057"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:27:20.165903: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 7338431921674375828\n",
      "2024-09-02 17:27:20.165933: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Epoch 10: val_loss did not improve from 1.84829\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 918us/step - accuracy: 0.5625 - loss: 1.8057 - val_accuracy: 0.5642 - val_loss: 1.8670 - learning_rate: 1.0000e-04\n",
      "Epoch 11/15\n",
      "\u001b[1m388/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.5753 - loss: 1.7924\n",
      "Epoch 11: val_loss improved from 1.84829 to 1.59269, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.5753 - loss: 1.7923 - val_accuracy: 0.6341 - val_loss: 1.5927 - learning_rate: 1.0000e-04\n",
      "Epoch 12/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 7ms/step - accuracy: 0.6328 - loss: 1.6477\n",
      "Epoch 12: val_loss improved from 1.59269 to 1.59072, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 1ms/step - accuracy: 0.6328 - loss: 1.6477 - val_accuracy: 0.6334 - val_loss: 1.5907 - learning_rate: 1.0000e-04\n",
      "Epoch 13/15\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.5940 - loss: 1.6835\n",
      "Epoch 13: val_loss improved from 1.59072 to 1.52057, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 22ms/step - accuracy: 0.5940 - loss: 1.6834 - val_accuracy: 0.6425 - val_loss: 1.5206 - learning_rate: 1.0000e-04\n",
      "Epoch 14/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 7ms/step - accuracy: 0.6484 - loss: 1.5705\n",
      "Epoch 14: val_loss did not improve from 1.52057\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 909us/step - accuracy: 0.6484 - loss: 1.5705 - val_accuracy: 0.6427 - val_loss: 1.5256 - learning_rate: 1.0000e-04\n",
      "Epoch 15/15\n",
      "\u001b[1m386/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.6195 - loss: 1.5780\n",
      "Epoch 15: val_loss improved from 1.52057 to 1.43229, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 22ms/step - accuracy: 0.6195 - loss: 1.5779 - val_accuracy: 0.6662 - val_loss: 1.4323 - learning_rate: 1.0000e-04\n",
      "Restoring model weights from the end of the best epoch: 15.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-09-02 17:27:46,988] Trial 1 finished with value: 0.6661999821662903 and parameters: {'lr': 0.00023575048409944037, 'batch_size': 128}. Best is trial 0 with value: 0.7577999830245972.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 26ms/step - accuracy: 0.2798 - loss: 3.1728\n",
      "Epoch 1: val_loss improved from inf to 4.08856, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m16s\u001b[0m 31ms/step - accuracy: 0.2799 - loss: 3.1718 - val_accuracy: 0.1999 - val_loss: 4.0886 - learning_rate: 1.0000e-04\n",
      "Epoch 2/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 7ms/step - accuracy: 0.4531 - loss: 2.3267\n",
      "Epoch 2: val_loss improved from 4.08856 to 4.07962, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 1ms/step - accuracy: 0.4531 - loss: 2.3267 - val_accuracy: 0.2005 - val_loss: 4.0796 - learning_rate: 1.0000e-04\n",
      "Epoch 3/15\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.4143 - loss: 2.4632\n",
      "Epoch 3: val_loss improved from 4.07962 to 2.08639, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.4144 - loss: 2.4631 - val_accuracy: 0.5383 - val_loss: 2.0864 - learning_rate: 1.0000e-04\n",
      "Epoch 4/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 7ms/step - accuracy: 0.5000 - loss: 2.1748\n",
      "Epoch 4: val_loss did not improve from 2.08639\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 910us/step - accuracy: 0.5000 - loss: 2.1748 - val_accuracy: 0.5393 - val_loss: 2.0871 - learning_rate: 1.0000e-04\n",
      "Epoch 5/15\n",
      "\u001b[1m386/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.4652 - loss: 2.2612\n",
      "Epoch 5: val_loss improved from 2.08639 to 1.99086, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.4653 - loss: 2.2607 - val_accuracy: 0.5550 - val_loss: 1.9909 - learning_rate: 1.0000e-04\n",
      "Epoch 6/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 7ms/step - accuracy: 0.5078 - loss: 2.0864\n",
      "Epoch 6: val_loss improved from 1.99086 to 1.98927, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 1ms/step - accuracy: 0.5078 - loss: 2.0864 - val_accuracy: 0.5570 - val_loss: 1.9893 - learning_rate: 1.0000e-04\n",
      "Epoch 7/15\n",
      "\u001b[1m386/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.5075 - loss: 2.0893\n",
      "Epoch 7: val_loss improved from 1.98927 to 1.90609, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.5077 - loss: 2.0887 - val_accuracy: 0.5722 - val_loss: 1.9061 - learning_rate: 1.0000e-04\n",
      "Epoch 8/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 6ms/step - accuracy: 0.4844 - loss: 2.0343\n",
      "Epoch 8: val_loss improved from 1.90609 to 1.90513, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 1ms/step - accuracy: 0.4844 - loss: 2.0343 - val_accuracy: 0.5713 - val_loss: 1.9051 - learning_rate: 1.0000e-04\n",
      "Epoch 9/15\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.5466 - loss: 1.9208\n",
      "Epoch 9: val_loss improved from 1.90513 to 1.82821, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.5466 - loss: 1.9207 - val_accuracy: 0.5855 - val_loss: 1.8282 - learning_rate: 1.0000e-04\n",
      "Epoch 10/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 7ms/step - accuracy: 0.5859 - loss: 1.8809\n",
      "Epoch 10: val_loss improved from 1.82821 to 1.82697, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 1ms/step - accuracy: 0.5859 - loss: 1.8809 - val_accuracy: 0.5857 - val_loss: 1.8270 - learning_rate: 1.0000e-04\n",
      "Epoch 11/15\n",
      "\u001b[1m385/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.5687 - loss: 1.7947\n",
      "Epoch 11: val_loss improved from 1.82697 to 1.59448, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 22ms/step - accuracy: 0.5688 - loss: 1.7942 - val_accuracy: 0.6384 - val_loss: 1.5945 - learning_rate: 1.0000e-04\n",
      "Epoch 12/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 7ms/step - accuracy: 0.6172 - loss: 1.7067\n",
      "Epoch 12: val_loss did not improve from 1.59448\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 962us/step - accuracy: 0.6172 - loss: 1.7067 - val_accuracy: 0.6365 - val_loss: 1.5965 - learning_rate: 1.0000e-04\n",
      "Epoch 13/15\n",
      "\u001b[1m388/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.5991 - loss: 1.6633\n",
      "Epoch 13: val_loss improved from 1.59448 to 1.51731, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.5991 - loss: 1.6631 - val_accuracy: 0.6481 - val_loss: 1.5173 - learning_rate: 1.0000e-04\n",
      "Epoch 14/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 7ms/step - accuracy: 0.6562 - loss: 1.3875\n",
      "Epoch 14: val_loss did not improve from 1.51731\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 933us/step - accuracy: 0.6562 - loss: 1.3875 - val_accuracy: 0.6471 - val_loss: 1.5220 - learning_rate: 1.0000e-04\n",
      "Epoch 15/15\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.6183 - loss: 1.5515\n",
      "Epoch 15: val_loss improved from 1.51731 to 1.46074, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.6183 - loss: 1.5515 - val_accuracy: 0.6516 - val_loss: 1.4607 - learning_rate: 1.0000e-04\n",
      "Restoring model weights from the end of the best epoch: 15.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-09-02 17:29:05,489] Trial 2 finished with value: 0.6516000032424927 and parameters: {'lr': 7.219761594323132e-05, 'batch_size': 128}. Best is trial 0 with value: 0.7577999830245972.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "\u001b[1m387/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 24ms/step - accuracy: 0.2938 - loss: 3.0998\n",
      "Epoch 1: val_loss improved from inf to 3.68033, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m15s\u001b[0m 30ms/step - accuracy: 0.2944 - loss: 3.0962 - val_accuracy: 0.2020 - val_loss: 3.6803 - learning_rate: 1.0000e-04\n",
      "Epoch 2/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m3s\u001b[0m 9ms/step - accuracy: 0.3906 - loss: 2.4333\n",
      "Epoch 2: val_loss did not improve from 3.68033\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 964us/step - accuracy: 0.3906 - loss: 2.4333 - val_accuracy: 0.2028 - val_loss: 3.7094 - learning_rate: 1.0000e-04\n",
      "Epoch 3/15\n",
      "\u001b[1m385/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.4269 - loss: 2.4331\n",
      "Epoch 3: val_loss improved from 3.68033 to 2.07517, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 21ms/step - accuracy: 0.4272 - loss: 2.4323 - val_accuracy: 0.5390 - val_loss: 2.0752 - learning_rate: 1.0000e-04\n",
      "Epoch 4/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 7ms/step - accuracy: 0.5000 - loss: 2.3285\n",
      "Epoch 4: val_loss did not improve from 2.07517\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 925us/step - accuracy: 0.5000 - loss: 2.3285 - val_accuracy: 0.5380 - val_loss: 2.0779 - learning_rate: 1.0000e-04\n",
      "Epoch 5/15\n",
      "\u001b[1m385/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.4825 - loss: 2.2298\n",
      "Epoch 5: val_loss improved from 2.07517 to 1.96909, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.4826 - loss: 2.2292 - val_accuracy: 0.5613 - val_loss: 1.9691 - learning_rate: 1.0000e-04\n",
      "Epoch 6/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 6ms/step - accuracy: 0.4844 - loss: 2.1476"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:29:38.581068: I tensorflow/core/framework/local_rendezvous.cc:404] Local rendezvous is aborting with status: OUT_OF_RANGE: End of sequence\n",
      "\t [[{{node IteratorGetNext}}]]\n",
      "\t [[IteratorGetNext/_2]]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Epoch 6: val_loss improved from 1.96909 to 1.94946, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 1ms/step - accuracy: 0.4844 - loss: 2.1476 - val_accuracy: 0.5651 - val_loss: 1.9495 - learning_rate: 1.0000e-04\n",
      "Epoch 7/15\n",
      "\u001b[1m385/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.5154 - loss: 2.0682\n",
      "Epoch 7: val_loss improved from 1.94946 to 1.78401, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 22ms/step - accuracy: 0.5156 - loss: 2.0676 - val_accuracy: 0.5970 - val_loss: 1.7840 - learning_rate: 1.0000e-04\n",
      "Epoch 8/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 6ms/step - accuracy: 0.4688 - loss: 1.9631\n",
      "Epoch 8: val_loss did not improve from 1.78401\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 917us/step - accuracy: 0.4688 - loss: 1.9631 - val_accuracy: 0.5948 - val_loss: 1.7980 - learning_rate: 1.0000e-04\n",
      "Epoch 9/15\n",
      "\u001b[1m385/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.5490 - loss: 1.9171\n",
      "Epoch 9: val_loss did not improve from 1.78401\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.5491 - loss: 1.9166 - val_accuracy: 0.5934 - val_loss: 1.8048 - learning_rate: 1.0000e-04\n",
      "Epoch 10/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 6ms/step - accuracy: 0.5391 - loss: 2.0109\n",
      "Epoch 10: val_loss did not improve from 1.78401\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 908us/step - accuracy: 0.5391 - loss: 2.0109 - val_accuracy: 0.5855 - val_loss: 1.8300 - learning_rate: 1.0000e-04\n",
      "Epoch 11/15\n",
      "\u001b[1m385/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.5730 - loss: 1.7910\n",
      "Epoch 11: val_loss improved from 1.78401 to 1.55375, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.5731 - loss: 1.7905 - val_accuracy: 0.6454 - val_loss: 1.5538 - learning_rate: 1.0000e-04\n",
      "Epoch 12/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 6ms/step - accuracy: 0.6484 - loss: 1.5696\n",
      "Epoch 12: val_loss did not improve from 1.55375\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 929us/step - accuracy: 0.6484 - loss: 1.5696 - val_accuracy: 0.6428 - val_loss: 1.5625 - learning_rate: 1.0000e-04\n",
      "Epoch 13/15\n",
      "\u001b[1m386/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.5934 - loss: 1.6732\n",
      "Epoch 13: val_loss improved from 1.55375 to 1.49138, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.5935 - loss: 1.6729 - val_accuracy: 0.6589 - val_loss: 1.4914 - learning_rate: 1.0000e-04\n",
      "Epoch 14/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 6ms/step - accuracy: 0.5234 - loss: 1.8179\n",
      "Epoch 14: val_loss did not improve from 1.49138\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 937us/step - accuracy: 0.5234 - loss: 1.8179 - val_accuracy: 0.6610 - val_loss: 1.4920 - learning_rate: 1.0000e-04\n",
      "Epoch 15/15\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.6151 - loss: 1.5705\n",
      "Epoch 15: val_loss improved from 1.49138 to 1.37457, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 21ms/step - accuracy: 0.6151 - loss: 1.5704 - val_accuracy: 0.6697 - val_loss: 1.3746 - learning_rate: 1.0000e-04\n",
      "Restoring model weights from the end of the best epoch: 15.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-09-02 17:30:22,841] Trial 3 finished with value: 0.669700026512146 and parameters: {'lr': 8.830156040787896e-05, 'batch_size': 128}. Best is trial 0 with value: 0.7577999830245972.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "\u001b[1m1556/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 7ms/step - accuracy: 0.3063 - loss: 2.9986\n",
      "Epoch 1: val_loss improved from inf to 2.39571, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m17s\u001b[0m 9ms/step - accuracy: 0.3065 - loss: 2.9971 - val_accuracy: 0.4541 - val_loss: 2.3957 - learning_rate: 1.0000e-04\n",
      "Epoch 2/15\n",
      "\u001b[1m   1/1562\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m7s\u001b[0m 5ms/step - accuracy: 0.4375 - loss: 2.4547\n",
      "Epoch 2: val_loss improved from 2.39571 to 2.38897, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 325us/step - accuracy: 0.4375 - loss: 2.4547 - val_accuracy: 0.4559 - val_loss: 2.3890 - learning_rate: 1.0000e-04\n",
      "Epoch 3/15\n",
      "\u001b[1m1556/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 6ms/step - accuracy: 0.4438 - loss: 2.2917\n",
      "Epoch 3: val_loss improved from 2.38897 to 1.96367, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m9s\u001b[0m 6ms/step - accuracy: 0.4438 - loss: 2.2913 - val_accuracy: 0.5414 - val_loss: 1.9637 - learning_rate: 1.0000e-04\n",
      "Epoch 4/15\n",
      "\u001b[1m   1/1562\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m6s\u001b[0m 4ms/step - accuracy: 0.4062 - loss: 1.9726\n",
      "Epoch 4: val_loss did not improve from 1.96367\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 240us/step - accuracy: 0.4062 - loss: 1.9726 - val_accuracy: 0.5394 - val_loss: 1.9653 - learning_rate: 1.0000e-04\n",
      "Epoch 5/15\n",
      "\u001b[1m1555/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 6ms/step - accuracy: 0.5090 - loss: 1.9624\n",
      "Epoch 5: val_loss improved from 1.96367 to 1.62406, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m10s\u001b[0m 6ms/step - accuracy: 0.5090 - loss: 1.9621 - val_accuracy: 0.5969 - val_loss: 1.6241 - learning_rate: 1.0000e-04\n",
      "Epoch 6/15\n",
      "\u001b[1m   1/1562\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m5s\u001b[0m 4ms/step - accuracy: 0.5312 - loss: 1.9090\n",
      "Epoch 6: val_loss improved from 1.62406 to 1.60980, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 342us/step - accuracy: 0.5312 - loss: 1.9090 - val_accuracy: 0.6003 - val_loss: 1.6098 - learning_rate: 1.0000e-04\n",
      "Epoch 7/15\n",
      "\u001b[1m1558/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 6ms/step - accuracy: 0.5623 - loss: 1.6811\n",
      "Epoch 7: val_loss improved from 1.60980 to 1.50684, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m10s\u001b[0m 6ms/step - accuracy: 0.5624 - loss: 1.6809 - val_accuracy: 0.5931 - val_loss: 1.5068 - learning_rate: 1.0000e-04\n",
      "Epoch 8/15\n",
      "\u001b[1m   1/1562\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m6s\u001b[0m 4ms/step - accuracy: 0.6562 - loss: 1.2332\n",
      "Epoch 8: val_loss improved from 1.50684 to 1.49137, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 330us/step - accuracy: 0.6562 - loss: 1.2332 - val_accuracy: 0.5978 - val_loss: 1.4914 - learning_rate: 1.0000e-04\n",
      "Epoch 9/15\n",
      "\u001b[1m1556/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 6ms/step - accuracy: 0.5991 - loss: 1.4716\n",
      "Epoch 9: val_loss improved from 1.49137 to 1.21955, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m10s\u001b[0m 6ms/step - accuracy: 0.5991 - loss: 1.4714 - val_accuracy: 0.6700 - val_loss: 1.2195 - learning_rate: 1.0000e-04\n",
      "Epoch 10/15\n",
      "\u001b[1m   1/1562\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m5s\u001b[0m 4ms/step - accuracy: 0.5938 - loss: 1.5289\n",
      "Epoch 10: val_loss did not improve from 1.21955\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 241us/step - accuracy: 0.5938 - loss: 1.5289 - val_accuracy: 0.6687 - val_loss: 1.2211 - learning_rate: 1.0000e-04\n",
      "Epoch 11/15\n",
      "\u001b[1m1555/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 6ms/step - accuracy: 0.6252 - loss: 1.3280\n",
      "Epoch 11: val_loss improved from 1.21955 to 1.08681, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m10s\u001b[0m 6ms/step - accuracy: 0.6253 - loss: 1.3279 - val_accuracy: 0.7049 - val_loss: 1.0868 - learning_rate: 1.0000e-04\n",
      "Epoch 12/15\n",
      "\u001b[1m   1/1562\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m6s\u001b[0m 4ms/step - accuracy: 0.5625 - loss: 1.4199\n",
      "Epoch 12: val_loss improved from 1.08681 to 1.08491, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 331us/step - accuracy: 0.5625 - loss: 1.4199 - val_accuracy: 0.7052 - val_loss: 1.0849 - learning_rate: 1.0000e-04\n",
      "Epoch 13/15\n",
      "\u001b[1m1554/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 6ms/step - accuracy: 0.6592 - loss: 1.2112\n",
      "Epoch 13: val_loss improved from 1.08491 to 1.07022, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m10s\u001b[0m 6ms/step - accuracy: 0.6592 - loss: 1.2111 - val_accuracy: 0.7049 - val_loss: 1.0702 - learning_rate: 1.0000e-04\n",
      "Epoch 14/15\n",
      "\u001b[1m   1/1562\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m5s\u001b[0m 4ms/step - accuracy: 0.6875 - loss: 1.0839\n",
      "Epoch 14: val_loss did not improve from 1.07022\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 239us/step - accuracy: 0.6875 - loss: 1.0839 - val_accuracy: 0.7016 - val_loss: 1.0778 - learning_rate: 1.0000e-04\n",
      "Epoch 15/15\n",
      "\u001b[1m1556/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 6ms/step - accuracy: 0.6856 - loss: 1.1147\n",
      "Epoch 15: val_loss improved from 1.07022 to 0.98094, saving model to best_model.keras\n",
      "\u001b[1m1562/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m10s\u001b[0m 6ms/step - accuracy: 0.6856 - loss: 1.1147 - val_accuracy: 0.7329 - val_loss: 0.9809 - learning_rate: 1.0000e-04\n",
      "Restoring model weights from the end of the best epoch: 15.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-09-02 17:31:51,038] Trial 4 finished with value: 0.7329000234603882 and parameters: {'lr': 1.0183854668236481e-05, 'batch_size': 32}. Best is trial 0 with value: 0.7577999830245972.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 24ms/step - accuracy: 0.2915 - loss: 3.1011\n",
      "Epoch 1: val_loss improved from inf to 3.85125, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m15s\u001b[0m 30ms/step - accuracy: 0.2917 - loss: 3.1002 - val_accuracy: 0.2009 - val_loss: 3.8512 - learning_rate: 1.0000e-04\n",
      "Epoch 2/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m3s\u001b[0m 9ms/step - accuracy: 0.3906 - loss: 2.5103\n",
      "Epoch 2: val_loss improved from 3.85125 to 3.84610, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 1ms/step - accuracy: 0.3906 - loss: 2.5103 - val_accuracy: 0.1995 - val_loss: 3.8461 - learning_rate: 1.0000e-04\n",
      "Epoch 3/15\n",
      "\u001b[1m386/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.4361 - loss: 2.4152\n",
      "Epoch 3: val_loss improved from 3.84610 to 2.14208, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.4362 - loss: 2.4145 - val_accuracy: 0.5236 - val_loss: 2.1421 - learning_rate: 1.0000e-04\n",
      "Epoch 4/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m3s\u001b[0m 8ms/step - accuracy: 0.5781 - loss: 2.1554\n",
      "Epoch 4: val_loss did not improve from 2.14208\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 954us/step - accuracy: 0.5781 - loss: 2.1554 - val_accuracy: 0.5228 - val_loss: 2.1482 - learning_rate: 1.0000e-04\n",
      "Epoch 5/15\n",
      "\u001b[1m386/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.4829 - loss: 2.2211\n",
      "Epoch 5: val_loss improved from 2.14208 to 2.01478, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.4830 - loss: 2.2206 - val_accuracy: 0.5275 - val_loss: 2.0148 - learning_rate: 1.0000e-04\n",
      "Epoch 6/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 6ms/step - accuracy: 0.4375 - loss: 2.2004\n",
      "Epoch 6: val_loss improved from 2.01478 to 2.00087, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 1ms/step - accuracy: 0.4375 - loss: 2.2004 - val_accuracy: 0.5314 - val_loss: 2.0009 - learning_rate: 1.0000e-04\n",
      "Epoch 7/15\n",
      "\u001b[1m388/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.5235 - loss: 2.0474\n",
      "Epoch 7: val_loss improved from 2.00087 to 1.90357, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.5235 - loss: 2.0471 - val_accuracy: 0.5733 - val_loss: 1.9036 - learning_rate: 1.0000e-04\n",
      "Epoch 8/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 7ms/step - accuracy: 0.5312 - loss: 2.0063\n",
      "Epoch 8: val_loss improved from 1.90357 to 1.87296, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 1ms/step - accuracy: 0.5312 - loss: 2.0063 - val_accuracy: 0.5774 - val_loss: 1.8730 - learning_rate: 1.0000e-04\n",
      "Epoch 9/15\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.5550 - loss: 1.8935\n",
      "Epoch 9: val_loss improved from 1.87296 to 1.58971, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.5550 - loss: 1.8934 - val_accuracy: 0.6432 - val_loss: 1.5897 - learning_rate: 1.0000e-04\n",
      "Epoch 10/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 7ms/step - accuracy: 0.5938 - loss: 1.8414\n",
      "Epoch 10: val_loss did not improve from 1.58971\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 933us/step - accuracy: 0.5938 - loss: 1.8414 - val_accuracy: 0.6414 - val_loss: 1.5937 - learning_rate: 1.0000e-04\n",
      "Epoch 11/15\n",
      "\u001b[1m386/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.5767 - loss: 1.7656\n",
      "Epoch 11: val_loss improved from 1.58971 to 1.49567, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.5768 - loss: 1.7652 - val_accuracy: 0.6547 - val_loss: 1.4957 - learning_rate: 1.0000e-04\n",
      "Epoch 12/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 7ms/step - accuracy: 0.6406 - loss: 1.6433\n",
      "Epoch 12: val_loss did not improve from 1.49567\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 959us/step - accuracy: 0.6406 - loss: 1.6433 - val_accuracy: 0.6571 - val_loss: 1.4962 - learning_rate: 1.0000e-04\n",
      "Epoch 13/15\n",
      "\u001b[1m386/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.5995 - loss: 1.6528\n",
      "Epoch 13: val_loss did not improve from 1.49567\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.5996 - loss: 1.6524 - val_accuracy: 0.6373 - val_loss: 1.5779 - learning_rate: 1.0000e-04\n",
      "Epoch 14/15\n",
      "\u001b[1m  1/390\u001b[0m \u001b[37m━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[1m2s\u001b[0m 6ms/step - accuracy: 0.5703 - loss: 1.5612"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 17:32:59.501796: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 7338431921674375828\n",
      "2024-09-02 17:32:59.501837: I tensorflow/core/framework/local_rendezvous.cc:423] Local rendezvous recv item cancelled. Key hash: 11929540167347183446\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Epoch 14: val_loss did not improve from 1.49567\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 930us/step - accuracy: 0.5703 - loss: 1.5612 - val_accuracy: 0.6329 - val_loss: 1.6087 - learning_rate: 1.0000e-04\n",
      "Epoch 15/15\n",
      "\u001b[1m386/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 20ms/step - accuracy: 0.6251 - loss: 1.5362\n",
      "Epoch 15: val_loss improved from 1.49567 to 1.30125, saving model to best_model.keras\n",
      "\u001b[1m390/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 21ms/step - accuracy: 0.6251 - loss: 1.5360 - val_accuracy: 0.6977 - val_loss: 1.3013 - learning_rate: 1.0000e-04\n",
      "Restoring model weights from the end of the best epoch: 15.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-09-02 17:33:08,263] Trial 5 finished with value: 0.697700023651123 and parameters: {'lr': 1.9942180076884915e-05, 'batch_size': 128}. Best is trial 0 with value: 0.7577999830245972.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "\u001b[1m775/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 12ms/step - accuracy: 0.3007 - loss: 3.0473\n",
      "Epoch 1: val_loss improved from inf to 2.17970, saving model to best_model.keras\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-09-02 17:33:26,324] Trial 6 pruned. Trial was pruned at epoch 0.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "\u001b[1m779/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 13ms/step - accuracy: 0.3021 - loss: 3.0518\n",
      "Epoch 1: val_loss improved from inf to 2.20636, saving model to best_model.keras\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-09-02 17:33:41,673] Trial 7 pruned. Trial was pruned at epoch 0.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "\u001b[1m388/390\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 25ms/step - accuracy: 0.2905 - loss: 3.1445\n",
      "Epoch 1: val_loss improved from inf to 3.44419, saving model to best_model.keras\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-09-02 17:33:57,107] Trial 8 pruned. Trial was pruned at epoch 0.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "\u001b[1m781/781\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 13ms/step - accuracy: 0.2998 - loss: 3.0483\n",
      "Epoch 1: val_loss improved from inf to 2.23639, saving model to best_model.keras\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-09-02 17:34:12,889] Trial 9 pruned. Trial was pruned at epoch 0.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "\u001b[1m1558/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 7ms/step - accuracy: 0.3033 - loss: 3.0017\n",
      "Epoch 1: val_loss improved from inf to 2.15404, saving model to best_model.keras\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-09-02 17:34:29,147] Trial 10 pruned. Trial was pruned at epoch 0.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "\u001b[1m1559/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 7ms/step - accuracy: 0.3078 - loss: 2.9610\n",
      "Epoch 1: val_loss improved from inf to 2.19538, saving model to best_model.keras\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-09-02 17:34:45,472] Trial 11 pruned. Trial was pruned at epoch 0.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "\u001b[1m1551/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 7ms/step - accuracy: 0.3063 - loss: 2.9831\n",
      "Epoch 1: val_loss improved from inf to 2.36782, saving model to best_model.keras\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-09-02 17:35:02,011] Trial 12 pruned. Trial was pruned at epoch 0.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "\u001b[1m1556/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 7ms/step - accuracy: 0.3056 - loss: 2.9855\n",
      "Epoch 1: val_loss improved from inf to 2.26686, saving model to best_model.keras\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-09-02 17:35:18,736] Trial 13 pruned. Trial was pruned at epoch 0.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "\u001b[1m1560/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 7ms/step - accuracy: 0.3036 - loss: 2.9851\n",
      "Epoch 1: val_loss improved from inf to 2.14429, saving model to best_model.keras\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-09-02 17:35:35,051] Trial 14 pruned. Trial was pruned at epoch 0.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "\u001b[1m1560/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 7ms/step - accuracy: 0.3024 - loss: 2.9791\n",
      "Epoch 1: val_loss improved from inf to 2.17978, saving model to best_model.keras\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-09-02 17:35:51,528] Trial 15 pruned. Trial was pruned at epoch 0.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "\u001b[1m1551/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 7ms/step - accuracy: 0.3078 - loss: 2.9672\n",
      "Epoch 1: val_loss improved from inf to 2.18227, saving model to best_model.keras\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-09-02 17:36:09,094] Trial 16 pruned. Trial was pruned at epoch 0.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "\u001b[1m1561/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 7ms/step - accuracy: 0.3082 - loss: 2.9874\n",
      "Epoch 1: val_loss improved from inf to 2.26275, saving model to best_model.keras\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-09-02 17:36:26,031] Trial 17 pruned. Trial was pruned at epoch 0.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "\u001b[1m1551/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 7ms/step - accuracy: 0.3093 - loss: 2.9718\n",
      "Epoch 1: val_loss improved from inf to 2.35094, saving model to best_model.keras\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-09-02 17:36:42,376] Trial 18 pruned. Trial was pruned at epoch 0.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/15\n",
      "\u001b[1m1557/1562\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━\u001b[0m \u001b[1m0s\u001b[0m 7ms/step - accuracy: 0.3055 - loss: 2.9599\n",
      "Epoch 1: val_loss improved from inf to 2.33576, saving model to best_model.keras\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[I 2024-09-02 17:36:58,813] Trial 19 pruned. Trial was pruned at epoch 0.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best trial:\n",
      "  Value: 0.7577999830245972\n",
      "  Params: \n",
      "    lr: 0.0008750803832793538\n",
      "    batch_size: 32\n"
     ]
    }
   ],
   "source": [
    "# cell2: Hyperparameter tuning using Optuna for optimizing the learning rate and batch size of the CNN model\n",
    "\n",
    "import optuna  # Importing Optuna for hyperparameter optimization\n",
    "from optuna.integration import KerasPruningCallback  # Importing KerasPruningCallback for pruning unpromising trials\n",
    "from keras.optimizers import Adam  # Importing Adam optimizer\n",
    "from tensorflow.keras.mixed_precision import set_global_policy  # Importing mixed precision policy for potential speedup\n",
    "from keras.callbacks import ReduceLROnPlateau, EarlyStopping, ModelCheckpoint  # Importing callbacks for learning rate reduction, early stopping, and model checkpointing\n",
    "\n",
    "# Enable mixed precision\n",
    "set_global_policy('mixed_float16')  # Enabling mixed precision to potentially speed up training and reduce memory usage\n",
    "\n",
    "# Define the objective function for Optuna\n",
    "def objective(trial):\n",
    "    # Hyperparameters to tune\n",
    "    lr = trial.suggest_float('lr', 1e-5, 1e-3, log=True)  # Suggesting a learning rate within a logarithmic range from 1e-5 to 1e-3\n",
    "    batch_size = trial.suggest_categorical('batch_size', [32, 64, 128])  # Suggesting a batch size from a list of options\n",
    "\n",
    "    # Use the existing model structure\n",
    "    model = Sequential([\n",
    "        # Same architecture as defined in your reference\n",
    "        Conv2D(64, (3, 3), padding='same', activation='relu', input_shape=(32, 32, 3)),  # First convolutional layer with 64 filters\n",
    "        BatchNormalization(),  # Batch normalization after the first convolutional layer\n",
    "        Conv2D(64, (3, 3), padding='same', activation='relu'),  # Second convolutional layer with 64 filters\n",
    "        BatchNormalization(),  # Batch normalization after the second convolutional layer\n",
    "        MaxPooling2D(pool_size=(2, 2)),  # Max pooling layer to reduce spatial dimensions\n",
    "        Conv2D(128, (3, 3), padding='same', activation='relu'),  # Third convolutional layer with 128 filters\n",
    "        BatchNormalization(),  # Batch normalization after the third convolutional layer\n",
    "        Conv2D(128, (3, 3), padding='same', activation='relu'),  # Fourth convolutional layer with 128 filters\n",
    "        BatchNormalization(),  # Batch normalization after the fourth convolutional layer\n",
    "        MaxPooling2D(pool_size=(2, 2)),  # Max pooling layer to reduce spatial dimensions\n",
    "        Conv2D(256, (3, 3), padding='same', activation='relu'),  # Fifth convolutional layer with 256 filters\n",
    "        BatchNormalization(),  # Batch normalization after the fifth convolutional layer\n",
    "        Conv2D(256, (3, 3), padding='same', activation='relu'),  # Sixth convolutional layer with 256 filters\n",
    "        BatchNormalization(),  # Batch normalization after the sixth convolutional layer\n",
    "        MaxPooling2D(pool_size=(2, 2)),  # Max pooling layer to reduce spatial dimensions\n",
    "        Flatten(),  # Flattening the output to feed into dense layers\n",
    "        Dense(512, activation='relu', kernel_regularizer=l2(0.001)),  # Dense layer with 512 units and L2 regularization\n",
    "        Dropout(0.5),  # Dropout layer for regularization\n",
    "        Dense(10, activation='softmax', dtype='float32')  # Output layer with softmax activation and float32 dtype to ensure compatibility with mixed precision\n",
    "    ])\n",
    "\n",
    "    # Compile model\n",
    "    model.compile(optimizer=Adam(learning_rate=lr), loss='categorical_crossentropy', metrics=['accuracy'])  # Compiling the model with the suggested learning rate\n",
    "\n",
    "    # Adjusted Learning Rate Scheduler\n",
    "    def lr_schedule(epoch):  # Defining a learning rate schedule function\n",
    "        lr = 1e-4  # Starting learning rate\n",
    "        if epoch > 50:  # If epoch number is greater than 50\n",
    "            lr *= 0.1  # Reduce learning rate by a factor of 0.1\n",
    "        elif epoch > 30:  # If epoch number is greater than 30 but less than or equal to 50\n",
    "            lr *= 0.5  # Reduce learning rate by a factor of 0.5\n",
    "        return lr  # Returning the adjusted learning rate\n",
    "\n",
    "    # Callbacks\n",
    "    lr_scheduler = LearningRateScheduler(lr_schedule)  # Creating a LearningRateScheduler callback\n",
    "    early_stopping = EarlyStopping(monitor='val_loss', patience=10, verbose=1, restore_best_weights=True)  # Creating an EarlyStopping callback\n",
    "    reduce_lr = ReduceLROnPlateau(monitor='val_loss', factor=0.5, patience=5, min_lr=1e-7, verbose=1)  # Creating a ReduceLROnPlateau callback\n",
    "    checkpoint = ModelCheckpoint('best_model.keras', monitor='val_loss', save_best_only=True, verbose=1)  # Creating a ModelCheckpoint callback to save the best model\n",
    "\n",
    "    # Train the model\n",
    "    history = model.fit(  # Training the model\n",
    "        datagen.flow(x_train, y_train, batch_size=batch_size),  # Using the data generator to generate augmented data with the suggested batch size\n",
    "        epochs=15,  # Number of epochs to train\n",
    "        validation_data=(x_val, y_val),  # Validation data\n",
    "        callbacks=[lr_scheduler, early_stopping, reduce_lr, checkpoint, KerasPruningCallback(trial, 'val_loss')],  # Using the defined callbacks including the pruning callback\n",
    "        steps_per_epoch=x_train.shape[0] // batch_size,  # Number of steps per epoch\n",
    "        verbose=1  # Verbosity level\n",
    "    )\n",
    "\n",
    "    # Get the best validation accuracy of this trial\n",
    "    val_accuracy = max(history.history['val_accuracy'])  # Extracting the best validation accuracy from the training history\n",
    "    return val_accuracy  # Returning the best validation accuracy for this trial\n",
    "\n",
    "\n",
    "# Run Optuna study\n",
    "study = optuna.create_study(direction='maximize')  # Creating an Optuna study with the objective to maximize validation accuracy\n",
    "study.optimize(objective, n_trials=20)  # Running the optimization with 20 trials\n",
    "\n",
    "print(\"Best trial:\")  # Printing the best trial information\n",
    "trial = study.best_trial  # Extracting the best trial\n",
    "print(f\"  Value: {trial.value}\")  # Printing the best validation accuracy\n",
    "print(\"  Params: \")  # Printing the best hyperparameters\n",
    "for key, value in trial.params.items():\n",
    "    print(f\"    {key}: {value}\")  # Printing each hyperparameter and its value"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/tmp/ipykernel_1015982/67270997.py:32: UserWarning: `input_shape` is undefined or non-square, or `rows` is not in [96, 128, 160, 192, 224]. Weights for input shape (224, 224) will be loaded as the default.\n",
      "  base_model = applications.MobileNetV2(weights='imagenet', include_top=False, input_shape=(112, 112, 3))\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m32s\u001b[0m 4ms/step - accuracy: 0.6043 - loss: 1.1685 - val_accuracy: 0.7660 - val_loss: 0.6962\n",
      "Epoch 2/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m24s\u001b[0m 4ms/step - accuracy: 0.7144 - loss: 0.8405 - val_accuracy: 0.7774 - val_loss: 0.6552\n",
      "Epoch 3/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m23s\u001b[0m 4ms/step - accuracy: 0.7406 - loss: 0.7754 - val_accuracy: 0.7788 - val_loss: 0.6629\n",
      "Epoch 4/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m24s\u001b[0m 4ms/step - accuracy: 0.7509 - loss: 0.7410 - val_accuracy: 0.7804 - val_loss: 0.6588\n",
      "Epoch 5/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m24s\u001b[0m 4ms/step - accuracy: 0.7555 - loss: 0.7214 - val_accuracy: 0.7869 - val_loss: 0.6479\n",
      "Epoch 6/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m23s\u001b[0m 4ms/step - accuracy: 0.7665 - loss: 0.6879 - val_accuracy: 0.7814 - val_loss: 0.6537\n",
      "Epoch 7/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m24s\u001b[0m 4ms/step - accuracy: 0.7672 - loss: 0.6838 - val_accuracy: 0.7891 - val_loss: 0.6366\n",
      "Epoch 8/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m24s\u001b[0m 4ms/step - accuracy: 0.7747 - loss: 0.6655 - val_accuracy: 0.7906 - val_loss: 0.6459\n",
      "Epoch 9/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m23s\u001b[0m 4ms/step - accuracy: 0.7715 - loss: 0.6607 - val_accuracy: 0.7899 - val_loss: 0.6453\n",
      "Epoch 10/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m23s\u001b[0m 4ms/step - accuracy: 0.7791 - loss: 0.6462 - val_accuracy: 0.7899 - val_loss: 0.6604\n",
      "Epoch 1/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m97s\u001b[0m 11ms/step - accuracy: 0.5514 - loss: 1.6741 - val_accuracy: 0.8088 - val_loss: 0.6016\n",
      "Epoch 2/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m65s\u001b[0m 10ms/step - accuracy: 0.7063 - loss: 0.9138 - val_accuracy: 0.8564 - val_loss: 0.4634\n",
      "Epoch 3/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m64s\u001b[0m 10ms/step - accuracy: 0.7678 - loss: 0.7317 - val_accuracy: 0.8799 - val_loss: 0.3899\n",
      "Epoch 4/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m64s\u001b[0m 10ms/step - accuracy: 0.8001 - loss: 0.6329 - val_accuracy: 0.8941 - val_loss: 0.3459\n",
      "Epoch 5/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m65s\u001b[0m 10ms/step - accuracy: 0.8257 - loss: 0.5529 - val_accuracy: 0.9015 - val_loss: 0.3241\n",
      "Epoch 6/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m65s\u001b[0m 10ms/step - accuracy: 0.8453 - loss: 0.4811 - val_accuracy: 0.9091 - val_loss: 0.2997\n",
      "Epoch 7/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m66s\u001b[0m 11ms/step - accuracy: 0.8626 - loss: 0.4309 - val_accuracy: 0.9104 - val_loss: 0.2911\n",
      "Epoch 8/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m472s\u001b[0m 76ms/step - accuracy: 0.8729 - loss: 0.3911 - val_accuracy: 0.9155 - val_loss: 0.2839\n",
      "Epoch 9/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m65s\u001b[0m 10ms/step - accuracy: 0.8901 - loss: 0.3575 - val_accuracy: 0.9174 - val_loss: 0.2778\n",
      "Epoch 10/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m66s\u001b[0m 11ms/step - accuracy: 0.8974 - loss: 0.3212 - val_accuracy: 0.9186 - val_loss: 0.2674\n"
     ]
    }
   ],
   "source": [
    "# cell3: Transfer learning using MobileNetV2 with reduced batch size and input size for efficient computation\n",
    "\n",
    "import tensorflow as tf  # Importing TensorFlow for deep learning\n",
    "from tensorflow.keras import layers, models, applications, optimizers  # Importing necessary modules from TensorFlow\n",
    "from tensorflow.keras.utils import to_categorical  # Importing utility for converting labels to one-hot encoding\n",
    "\n",
    "# Load CIFAR-10 dataset\n",
    "(x_train, y_train), (x_val, y_val) = tf.keras.datasets.cifar10.load_data()  # Loading the CIFAR-10 dataset\n",
    "\n",
    "# Normalize pixel values to be between 0 and 1\n",
    "x_train, x_val = x_train / 255.0, x_val / 255.0  # Normalizing the pixel values to the range [0, 1]\n",
    "\n",
    "# Convert labels to one-hot encoding\n",
    "y_train = to_categorical(y_train, 10)  # Converting training labels to one-hot encoding\n",
    "y_val = to_categorical(y_val, 10)  # Converting validation labels to one-hot encoding\n",
    "\n",
    "# Parameters\n",
    "batch_size = 8  # Further reduced batch size to reduce memory usage and potentially improve convergence\n",
    "input_shape = (112, 112)  # Smaller input size to reduce computational load\n",
    "\n",
    "# Create a tf.data.Dataset for efficient loading\n",
    "train_dataset = tf.data.Dataset.from_tensor_slices((x_train, y_train))  # Creating a dataset from the training data\n",
    "train_dataset = train_dataset.map(lambda x, y: (tf.image.resize(x, input_shape), y), num_parallel_calls=tf.data.AUTOTUNE)  # Resizing images to the specified input shape\n",
    "train_dataset = train_dataset.shuffle(buffer_size=10000).batch(batch_size).prefetch(tf.data.AUTOTUNE)  # Shuffling, batching, and prefetching for efficient data loading\n",
    "\n",
    "val_dataset = tf.data.Dataset.from_tensor_slices((x_val, y_val))  # Creating a dataset from the validation data\n",
    "val_dataset = val_dataset.map(lambda x, y: (tf.image.resize(x, input_shape), y), num_parallel_calls=tf.data.AUTOTUNE)  # Resizing images to the specified input shape\n",
    "val_dataset = val_dataset.batch(batch_size).prefetch(tf.data.AUTOTUNE)  # Batching and prefetching for efficient data loading\n",
    "\n",
    "# Define the number of classes from your dataset\n",
    "num_classes = y_train.shape[-1]  # Extracting the number of classes from the training labels\n",
    "\n",
    "# Load a lighter pre-trained model (e.g., MobileNetV2) without the top classification layer\n",
    "base_model = applications.MobileNetV2(weights='imagenet', include_top=False, input_shape=(112, 112, 3))  # Loading MobileNetV2 with pre-trained weights and excluding the top classification layer\n",
    "base_model.trainable = False  # Freezing the base model layers to prevent further training\n",
    "\n",
    "# Add custom top layers\n",
    "model = models.Sequential([\n",
    "    base_model,  # Adding the base model\n",
    "    layers.GlobalAveragePooling2D(),  # Adding a global average pooling layer to reduce spatial dimensions\n",
    "    layers.Dense(128, activation='relu'),  # Adding a dense layer with 128 units and ReLU activation\n",
    "    layers.Dropout(0.5),  # Adding a dropout layer for regularization\n",
    "    layers.Dense(num_classes, activation='softmax')  # Adding the output layer with softmax activation\n",
    "])\n",
    "\n",
    "# Compile the model\n",
    "model.compile(\n",
    "    optimizer=optimizers.Adam(learning_rate=1e-3),  # Compiling the model with Adam optimizer and a learning rate of 1e-3\n",
    "    loss='categorical_crossentropy',  # Using categorical cross-entropy loss\n",
    "    metrics=['accuracy']  # Monitoring accuracy during training\n",
    ")\n",
    "\n",
    "# Training the model with frozen base layers\n",
    "history = model.fit(\n",
    "    train_dataset,  # Training dataset\n",
    "    validation_data=val_dataset,  # Validation dataset\n",
    "    epochs=10,  # Number of epochs to train\n",
    "    callbacks=[tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=3, restore_best_weights=True)],  # Early stopping callback\n",
    "    verbose=1  # Verbosity level\n",
    ")\n",
    "\n",
    "# Fine-tune by unfreezing some layers in the base model\n",
    "base_model.trainable = True  # Unfreezing the base model layers for fine-tuning\n",
    "\n",
    "# Compile the model again with a lower learning rate\n",
    "model.compile(\n",
    "    optimizer=optimizers.Adam(learning_rate=1e-5),  # Compiling the model with a lower learning rate for fine-tuning\n",
    "    loss='categorical_crossentropy',  # Using categorical cross-entropy loss\n",
    "    metrics=['accuracy']  # Monitoring accuracy during training\n",
    ")\n",
    "\n",
    "# Fine-tuning the model\n",
    "fine_tune_history = model.fit(\n",
    "    train_dataset,  # Training dataset\n",
    "    validation_data=val_dataset,  # Validation dataset\n",
    "    epochs=10,  # Number of epochs to fine-tune\n",
    "    callbacks=[tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=3, restore_best_weights=True)],  # Early stopping callback\n",
    "    verbose=1  # Verbosity level\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/tmp/ipykernel_1015982/2543941803.py:34: UserWarning: `input_shape` is undefined or non-square, or `rows` is not in [96, 128, 160, 192, 224]. Weights for input shape (224, 224) will be loaded as the default.\n",
      "  base_model = applications.MobileNetV2(weights='imagenet', include_top=False, input_shape=(112, 112, 3))\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m35s\u001b[0m 5ms/step - accuracy: 0.6063 - loss: 1.1741 - val_accuracy: 0.7629 - val_loss: 0.6874\n",
      "Epoch 2/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m29s\u001b[0m 5ms/step - accuracy: 0.7200 - loss: 0.8306 - val_accuracy: 0.7770 - val_loss: 0.6644\n",
      "Epoch 3/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m31s\u001b[0m 5ms/step - accuracy: 0.7387 - loss: 0.7714 - val_accuracy: 0.7819 - val_loss: 0.6420\n",
      "Epoch 4/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m28s\u001b[0m 4ms/step - accuracy: 0.7529 - loss: 0.7296 - val_accuracy: 0.7879 - val_loss: 0.6404\n",
      "Epoch 5/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m28s\u001b[0m 4ms/step - accuracy: 0.7570 - loss: 0.7150 - val_accuracy: 0.7800 - val_loss: 0.6625\n",
      "Epoch 6/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m28s\u001b[0m 4ms/step - accuracy: 0.7611 - loss: 0.6898 - val_accuracy: 0.7845 - val_loss: 0.6550\n",
      "Epoch 7/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m27s\u001b[0m 4ms/step - accuracy: 0.7629 - loss: 0.6843 - val_accuracy: 0.7853 - val_loss: 0.6464\n",
      "Epoch 1/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m89s\u001b[0m 11ms/step - accuracy: 0.5602 - loss: 1.4688 - val_accuracy: 0.8059 - val_loss: 0.5976\n",
      "Epoch 2/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m70s\u001b[0m 11ms/step - accuracy: 0.7150 - loss: 0.8994 - val_accuracy: 0.8559 - val_loss: 0.4545\n",
      "Epoch 3/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m71s\u001b[0m 11ms/step - accuracy: 0.7690 - loss: 0.7281 - val_accuracy: 0.8817 - val_loss: 0.3790\n",
      "Epoch 4/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m70s\u001b[0m 11ms/step - accuracy: 0.8052 - loss: 0.6255 - val_accuracy: 0.8920 - val_loss: 0.3419\n",
      "Epoch 5/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m71s\u001b[0m 11ms/step - accuracy: 0.8284 - loss: 0.5451 - val_accuracy: 0.9014 - val_loss: 0.3155\n",
      "Epoch 6/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m71s\u001b[0m 11ms/step - accuracy: 0.8499 - loss: 0.4773 - val_accuracy: 0.9072 - val_loss: 0.2927\n",
      "Epoch 7/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m71s\u001b[0m 11ms/step - accuracy: 0.8632 - loss: 0.4209 - val_accuracy: 0.9180 - val_loss: 0.2710\n",
      "Epoch 8/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m71s\u001b[0m 11ms/step - accuracy: 0.8761 - loss: 0.3846 - val_accuracy: 0.9195 - val_loss: 0.2616\n",
      "Epoch 9/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m70s\u001b[0m 11ms/step - accuracy: 0.8846 - loss: 0.3562 - val_accuracy: 0.9189 - val_loss: 0.2720\n",
      "Epoch 10/10\n",
      "\u001b[1m6250/6250\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m71s\u001b[0m 11ms/step - accuracy: 0.8963 - loss: 0.3278 - val_accuracy: 0.9213 - val_loss: 0.2644\n"
     ]
    }
   ],
   "source": [
    "# cell4: Transfer learning using MobileNetV2 with TensorBoard logging for visualization and analysis\n",
    "\n",
    "import datetime  # Importing datetime for logging purposes\n",
    "import tensorflow as tf  # Importing TensorFlow for deep learning\n",
    "from tensorflow.keras import layers, models, applications, optimizers  # Importing necessary modules from TensorFlow\n",
    "from tensorflow.keras.utils import to_categorical  # Importing utility for converting labels to one-hot encoding\n",
    "from tensorflow.keras.callbacks import TensorBoard  # Importing TensorBoard for logging and visualization\n",
    "\n",
    "# Load CIFAR-10 dataset\n",
    "(x_train, y_train), (x_val, y_val) = tf.keras.datasets.cifar10.load_data()  # Loading the CIFAR-10 dataset\n",
    "\n",
    "# Normalize pixel values to be between 0 and 1\n",
    "x_train, x_val = x_train / 255.0, x_val / 255.0  # Normalizing the pixel values to the range [0, 1]\n",
    "\n",
    "# Convert labels to one-hot encoding\n",
    "y_train = to_categorical(y_train, 10)  # Converting training labels to one-hot encoding\n",
    "y_val = to_categorical(y_val, 10)  # Converting validation labels to one-hot encoding\n",
    "\n",
    "# Parameters\n",
    "batch_size = 8  # Further reduced batch size to reduce memory usage and potentially improve convergence\n",
    "input_shape = (112, 112)  # Smaller input size to reduce computational load\n",
    "\n",
    "# Create a tf.data.Dataset for efficient loading\n",
    "train_dataset = tf.data.Dataset.from_tensor_slices((x_train, y_train))  # Creating a dataset from the training data\n",
    "train_dataset = train_dataset.map(lambda x, y: (tf.image.resize(x, input_shape), y), num_parallel_calls=tf.data.AUTOTUNE)  # Resizing images to the specified input shape\n",
    "train_dataset = train_dataset.shuffle(buffer_size=10000).batch(batch_size).prefetch(tf.data.AUTOTUNE)  # Shuffling, batching, and prefetching for efficient data loading\n",
    "\n",
    "val_dataset = tf.data.Dataset.from_tensor_slices((x_val, y_val))  # Creating a dataset from the validation data\n",
    "val_dataset = val_dataset.map(lambda x, y: (tf.image.resize(x, input_shape), y), num_parallel_calls=tf.data.AUTOTUNE)  # Resizing images to the specified input shape\n",
    "val_dataset = val_dataset.batch(batch_size).prefetch(tf.data.AUTOTUNE)  # Batching and prefetching for efficient data loading\n",
    "\n",
    "# Define the number of classes from your dataset\n",
    "num_classes = y_train.shape[-1]  # Extracting the number of classes from the training labels\n",
    "\n",
    "# Load a lighter pre-trained model (e.g., MobileNetV2) without the top classification layer\n",
    "base_model = applications.MobileNetV2(weights='imagenet', include_top=False, input_shape=(112, 112, 3))  # Loading MobileNetV2 with pre-trained weights and excluding the top classification layer\n",
    "base_model.trainable = False  # Freezing the base model layers to prevent further training\n",
    "\n",
    "# Add custom top layers\n",
    "model = models.Sequential([\n",
    "    base_model,  # Adding the base model\n",
    "    layers.GlobalAveragePooling2D(),  # Adding a global average pooling layer to reduce spatial dimensions\n",
    "    layers.Dense(128, activation='relu'),  # Adding a dense layer with 128 units and ReLU activation\n",
    "    layers.Dropout(0.5),  # Adding a dropout layer for regularization\n",
    "    layers.Dense(num_classes, activation='softmax')  # Adding the output layer with softmax activation\n",
    "])\n",
    "\n",
    "# Compile the model\n",
    "model.compile(\n",
    "    optimizer=optimizers.Adam(learning_rate=1e-3),  # Compiling the model with Adam optimizer and a learning rate of 1e-3\n",
    "    loss='categorical_crossentropy',  # Using categorical cross-entropy loss\n",
    "    metrics=['accuracy']  # Monitoring accuracy during training\n",
    ")\n",
    "\n",
    "# Define the TensorBoard callback\n",
    "log_dir = \"logs/fit/\" + datetime.datetime.now().strftime(\"%Y%m%d-%H%M%S\")  # Defining the log directory for TensorBoard\n",
    "tensorboard_callback = TensorBoard(log_dir=log_dir, histogram_freq=1)  # Creating a TensorBoard callback for logging\n",
    "\n",
    "# Training the model with frozen base layers\n",
    "history = model.fit(\n",
    "    train_dataset,  # Training dataset\n",
    "    validation_data=val_dataset,  # Validation dataset\n",
    "    epochs=10,  # Number of epochs to train\n",
    "    callbacks=[\n",
    "        tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=3, restore_best_weights=True),  # Early stopping callback\n",
    "        tensorboard_callback  # TensorBoard callback for logging\n",
    "    ],\n",
    "    verbose=1  # Verbosity level\n",
    ")\n",
    "\n",
    "# Fine-tune by unfreezing some layers in the base model\n",
    "base_model.trainable = True  # Unfreezing the base model layers for fine-tuning\n",
    "\n",
    "# Compile the model again with a lower learning rate\n",
    "model.compile(\n",
    "    optimizer=optimizers.Adam(learning_rate=1e-5),  # Compiling the model with a lower learning rate for fine-tuning\n",
    "    loss='categorical_crossentropy',  # Using categorical cross-entropy loss\n",
    "    metrics=['accuracy']  # Monitoring accuracy during training\n",
    ")\n",
    "\n",
    "# Fine-tuning the model\n",
    "fine_tune_history = model.fit(\n",
    "    train_dataset,  # Training dataset\n",
    "    validation_data=val_dataset,  # Validation dataset\n",
    "    epochs=10,  # Number of epochs to fine-tune\n",
    "    callbacks=[\n",
    "        tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=3, restore_best_weights=True),  # Early stopping callback\n",
    "        tensorboard_callback  # TensorBoard callback for logging\n",
    "    ],\n",
    "    verbose=1  # Verbosity level\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fold 1\n",
      "Epoch 1/10\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/matt/CIFAR10_CNN_Project/venv/lib/python3.12/site-packages/keras/src/trainers/data_adapters/py_dataset_adapter.py:122: UserWarning: Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored.\n",
      "  self._warn_if_super_not_called()\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m6s\u001b[0m 115ms/step - accuracy: 0.1044 - loss: 2.5379 - val_accuracy: 0.1520 - val_loss: 2.2940 - learning_rate: 1.0000e-05\n",
      "Epoch 2/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 61ms/step - accuracy: 0.1602 - loss: 2.2780 - val_accuracy: 0.2100 - val_loss: 2.2109 - learning_rate: 1.0000e-05\n",
      "Epoch 3/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 59ms/step - accuracy: 0.2252 - loss: 2.1867 - val_accuracy: 0.2940 - val_loss: 2.1318 - learning_rate: 1.0000e-05\n",
      "Epoch 4/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 60ms/step - accuracy: 0.2843 - loss: 2.0885 - val_accuracy: 0.3060 - val_loss: 2.0584 - learning_rate: 1.0000e-05\n",
      "Epoch 5/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 61ms/step - accuracy: 0.3527 - loss: 1.9862 - val_accuracy: 0.3500 - val_loss: 1.9706 - learning_rate: 1.0000e-05\n",
      "Epoch 6/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 60ms/step - accuracy: 0.3679 - loss: 1.8997 - val_accuracy: 0.3500 - val_loss: 1.8857 - learning_rate: 9.0484e-06\n",
      "Epoch 7/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 59ms/step - accuracy: 0.4124 - loss: 1.7398 - val_accuracy: 0.3920 - val_loss: 1.8304 - learning_rate: 8.1873e-06\n",
      "Epoch 8/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 61ms/step - accuracy: 0.4730 - loss: 1.6785 - val_accuracy: 0.3980 - val_loss: 1.7919 - learning_rate: 7.4082e-06\n",
      "Epoch 9/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 59ms/step - accuracy: 0.4550 - loss: 1.6396 - val_accuracy: 0.4080 - val_loss: 1.7397 - learning_rate: 6.7032e-06\n",
      "Epoch 10/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 61ms/step - accuracy: 0.4756 - loss: 1.5205 - val_accuracy: 0.4080 - val_loss: 1.6975 - learning_rate: 6.0653e-06\n",
      "Fold 2\n",
      "Epoch 1/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 64ms/step - accuracy: 0.4065 - loss: 1.7212 - val_accuracy: 0.5340 - val_loss: 1.4363 - learning_rate: 6.0653e-06\n",
      "Epoch 2/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 61ms/step - accuracy: 0.4398 - loss: 1.5809 - val_accuracy: 0.5080 - val_loss: 1.4368 - learning_rate: 6.0653e-06\n",
      "Epoch 3/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 60ms/step - accuracy: 0.4526 - loss: 1.5616 - val_accuracy: 0.5500 - val_loss: 1.4000 - learning_rate: 6.0653e-06\n",
      "Epoch 4/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 61ms/step - accuracy: 0.4561 - loss: 1.5295 - val_accuracy: 0.5320 - val_loss: 1.3873 - learning_rate: 6.0653e-06\n",
      "Epoch 5/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 60ms/step - accuracy: 0.4473 - loss: 1.5119 - val_accuracy: 0.5100 - val_loss: 1.3967 - learning_rate: 6.0653e-06\n",
      "Epoch 6/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 60ms/step - accuracy: 0.5076 - loss: 1.4499 - val_accuracy: 0.5220 - val_loss: 1.3642 - learning_rate: 5.4881e-06\n",
      "Epoch 7/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 59ms/step - accuracy: 0.5330 - loss: 1.4349 - val_accuracy: 0.5160 - val_loss: 1.3703 - learning_rate: 4.9659e-06\n",
      "Epoch 8/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 60ms/step - accuracy: 0.5128 - loss: 1.4355 - val_accuracy: 0.5140 - val_loss: 1.3607 - learning_rate: 4.4933e-06\n",
      "Epoch 9/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 61ms/step - accuracy: 0.5371 - loss: 1.3385 - val_accuracy: 0.5020 - val_loss: 1.3458 - learning_rate: 4.0657e-06\n",
      "Epoch 10/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 62ms/step - accuracy: 0.5653 - loss: 1.3331 - val_accuracy: 0.5260 - val_loss: 1.3244 - learning_rate: 3.6788e-06\n",
      "Test Accuracy: 0.46000000834465027\n",
      "\u001b[1m7/7\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 59ms/step\n",
      "Precision: 0.48210802742149483\n",
      "Recall: 0.46\n",
      "F1-Score: 0.45696542882981744\n",
      "ROC-AUC: 0.8721693513561938\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1000x700 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# cell5: Transfer learning using VGG16 with K-Fold Cross-Validation for robust evaluation\n",
    "\n",
    "import numpy as np  # Importing numpy for numerical operations\n",
    "import tensorflow as tf  # Importing TensorFlow for deep learning\n",
    "from tensorflow.keras.applications import VGG16  # Importing VGG16 pre-trained model\n",
    "from tensorflow.keras.models import Model  # Importing Model class for creating a model\n",
    "from tensorflow.keras.layers import Dense, GlobalAveragePooling2D  # Importing layers for building the model\n",
    "from tensorflow.keras.callbacks import LearningRateScheduler, EarlyStopping  # Importing callbacks for learning rate scheduling and early stopping\n",
    "from tensorflow.keras.preprocessing.image import ImageDataGenerator  # Importing ImageDataGenerator for data augmentation\n",
    "from sklearn.model_selection import KFold  # Importing KFold for cross-validation\n",
    "from sklearn.metrics import confusion_matrix, precision_score, recall_score, f1_score, roc_auc_score  # Importing metrics for evaluation\n",
    "import matplotlib.pyplot as plt  # Importing matplotlib for plotting\n",
    "import seaborn as sns  # Importing seaborn for enhanced visualizations\n",
    "\n",
    "# Load and preprocess data\n",
    "(x_train, y_train), (x_test, y_test) = tf.keras.datasets.cifar10.load_data()  # Loading the CIFAR-10 dataset\n",
    "x_train, x_test = x_train / 255.0, x_test / 255.0  # Normalizing the pixel values to the range [0, 1]\n",
    "y_train = y_train.flatten()  # Flattening the training labels\n",
    "y_test = y_test.flatten()  # Flattening the test labels\n",
    "\n",
    "# Use a smaller subset of the data for testing\n",
    "x_train_small = x_train[:1000]  # Using the first 1000 training images\n",
    "y_train_small = y_train[:1000]  # Using the first 1000 training labels\n",
    "x_test_small = x_test[:200]  # Using the first 200 test images\n",
    "y_test_small = y_test[:200]  # Using the first 200 test labels\n",
    "\n",
    "# Resize images to (224, 224) for VGG16\n",
    "x_train_small = np.array([tf.image.resize(img, [224, 224]) for img in x_train_small])  # Resizing training images to (224, 224)\n",
    "x_test_small = np.array([tf.image.resize(img, [224, 224]) for img in x_test_small])  # Resizing test images to (224, 224)\n",
    "\n",
    "# Data Augmentation\n",
    "datagen = ImageDataGenerator(  # Creating an ImageDataGenerator for data augmentation\n",
    "    width_shift_range=0.1,  # Allowing horizontal shifts\n",
    "    height_shift_range=0.1,  # Allowing vertical shifts\n",
    "    horizontal_flip=True  # Allowing horizontal flips\n",
    ")\n",
    "datagen.fit(x_train_small)  # Fitting the data generator on the training data\n",
    "\n",
    "# Load pre-trained VGG16 model and add custom layers\n",
    "base_model = VGG16(weights='imagenet', include_top=False, input_shape=(224, 224, 3))  # Loading VGG16 with pre-trained weights and excluding the top classification layer\n",
    "x = base_model.output  # Getting the output of the base model\n",
    "x = GlobalAveragePooling2D()(x)  # Adding a global average pooling layer to reduce spatial dimensions\n",
    "x = Dense(128, activation='relu')(x)  # Adding a dense layer with 128 units and ReLU activation\n",
    "predictions = Dense(10, activation='softmax')(x)  # Adding the output layer with softmax activation\n",
    "\n",
    "model = Model(inputs=base_model.input, outputs=predictions)  # Creating the model\n",
    "\n",
    "# Freeze the base model layers\n",
    "for layer in base_model.layers:  # Iterating over the layers of the base model\n",
    "    layer.trainable = False  # Freezing the layers to prevent further training\n",
    "\n",
    "# Unfreeze the top layers\n",
    "for layer in base_model.layers[-4:]:  # Iterating over the last 4 layers of the base model\n",
    "    layer.trainable = True  # Unfreezing the layers for fine-tuning\n",
    "\n",
    "# Compile the model\n",
    "model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=1e-5), loss='sparse_categorical_crossentropy', metrics=['accuracy'])  # Compiling the model with Adam optimizer and sparse categorical cross-entropy loss\n",
    "\n",
    "# Learning Rate Scheduler\n",
    "def lr_scheduler(epoch, lr):  # Defining a learning rate schedule function\n",
    "    if epoch < 5:  # If epoch number is less than 5\n",
    "        return lr  # Return the current learning rate\n",
    "    else:  # Otherwise\n",
    "        return float(lr * tf.math.exp(-0.1))  # Reduce the learning rate by a factor of exp(-0.1)\n",
    "\n",
    "# Early Stopping\n",
    "early_stopping = EarlyStopping(monitor='val_loss', patience=5, restore_best_weights=True)  # Creating an EarlyStopping callback\n",
    "\n",
    "# K-Fold Cross-Validation\n",
    "kfold = KFold(n_splits=2, shuffle=True, random_state=42)  # Creating a KFold cross-validator with 2 splits\n",
    "fold_no = 1  # Initializing the fold number\n",
    "history = []  # Initializing a list to store the training history\n",
    "\n",
    "for train, val in kfold.split(x_train_small, y_train_small):  # Iterating over the folds\n",
    "    print(f'Fold {fold_no}')  # Printing the current fold number\n",
    "    \n",
    "    # Train the model with reduced batch size and fewer epochs\n",
    "    history.append(model.fit(  # Training the model\n",
    "        datagen.flow(x_train_small[train], y_train_small[train], batch_size=16),  # Using the data generator to generate augmented data with a batch size of 16\n",
    "        epochs=10,  # Number of epochs to train\n",
    "        validation_data=(x_train_small[val], y_train_small[val]),  # Validation data\n",
    "        callbacks=[LearningRateScheduler(lr_scheduler), early_stopping],  # Using the defined callbacks\n",
    "        verbose=1  # Verbosity level\n",
    "    ))\n",
    "    \n",
    "    fold_no += 1  # Incrementing the fold number\n",
    "\n",
    "# Evaluate model on test set\n",
    "test_loss, test_accuracy = model.evaluate(x_test_small, y_test_small, verbose=0)  # Evaluating the model on the test set\n",
    "print(f'Test Accuracy: {test_accuracy}')  # Printing the test accuracy\n",
    "\n",
    "# Calculate additional metrics\n",
    "y_pred = model.predict(x_test_small)  # Predicting on the test set\n",
    "y_pred_classes = np.argmax(y_pred, axis=1)  # Getting the predicted classes\n",
    "\n",
    "conf_matrix = confusion_matrix(y_test_small, y_pred_classes)  # Calculating the confusion matrix\n",
    "precision = precision_score(y_test_small, y_pred_classes, average='weighted')  # Calculating the precision\n",
    "recall = recall_score(y_test_small, y_pred_classes, average='weighted')  # Calculating the recall\n",
    "f1 = f1_score(y_test_small, y_pred_classes, average='weighted')  # Calculating the F1-score\n",
    "\n",
    "# Ensure y_pred are probabilities\n",
    "y_pred_prob = tf.nn.softmax(y_pred, axis=1).numpy()  # Converting the predictions to probabilities\n",
    "roc_auc = roc_auc_score(y_test_small, y_pred_prob, multi_class='ovr')  # Calculating the ROC-AUC score\n",
    "\n",
    "print(f'Precision: {precision}')  # Printing the precision\n",
    "print(f'Recall: {recall}')  # Printing the recall\n",
    "print(f'F1-Score: {f1}')  # Printing the F1-score\n",
    "print(f'ROC-AUC: {roc_auc}')  # Printing the ROC-AUC score\n",
    "\n",
    "# Visualize confusion matrix\n",
    "plt.figure(figsize=(10, 7))  # Creating a figure for plotting\n",
    "sns.heatmap(conf_matrix, annot=True, fmt='d', cmap='Blues', xticklabels=range(10), yticklabels=range(10))  # Plotting the confusion matrix\n",
    "plt.xlabel('Predicted')  # Setting the x-axis label\n",
    "plt.ylabel('True')  # Setting the y-axis label\n",
    "plt.title('Confusion Matrix')  # Setting the title\n",
    "plt.show()  # Displaying the plot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fold 1\n",
      "Epoch 1/10\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/matt/CIFAR10_CNN_Project/venv/lib/python3.12/site-packages/keras/src/trainers/data_adapters/py_dataset_adapter.py:122: UserWarning: Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored.\n",
      "  self._warn_if_super_not_called()\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m25/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━\u001b[0m\u001b[37m━━━━━\u001b[0m \u001b[1m1s\u001b[0m 192ms/step - accuracy: 0.0610 - loss: 2.5585"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 19:49:37.261198: I external/local_xla/xla/stream_executor/cuda/cuda_asm_compiler.cc:393] ptxas warning : Registers are spilled to local memory in function 'gemm_fusion_dot_1169', 148 bytes spill stores, 148 bytes spill loads\n",
      "\n",
      "2024-09-02 19:49:40.934789: I external/local_xla/xla/stream_executor/cuda/cuda_asm_compiler.cc:393] ptxas warning : Registers are spilled to local memory in function 'gemm_fusion_dot_1169', 12 bytes spill stores, 12 bytes spill loads\n",
      "\n",
      "2024-09-02 19:49:40.993184: I external/local_xla/xla/stream_executor/cuda/cuda_asm_compiler.cc:393] ptxas warning : Registers are spilled to local memory in function 'gemm_fusion_dot_1169', 192 bytes spill stores, 192 bytes spill loads\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m19s\u001b[0m 389ms/step - accuracy: 0.0636 - loss: 2.5439 - val_accuracy: 0.0920 - val_loss: 2.7549 - learning_rate: 1.0000e-05\n",
      "Epoch 2/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 53ms/step - accuracy: 0.1397 - loss: 2.3668 - val_accuracy: 0.1100 - val_loss: 2.5459 - learning_rate: 1.0000e-05\n",
      "Epoch 3/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 49ms/step - accuracy: 0.1536 - loss: 2.2000 - val_accuracy: 0.1320 - val_loss: 2.3781 - learning_rate: 1.0000e-05\n",
      "Epoch 4/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 47ms/step - accuracy: 0.2253 - loss: 2.1189 - val_accuracy: 0.1720 - val_loss: 2.2356 - learning_rate: 1.0000e-05\n",
      "Epoch 5/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 50ms/step - accuracy: 0.3022 - loss: 1.9952 - val_accuracy: 0.2240 - val_loss: 2.1165 - learning_rate: 1.0000e-05\n",
      "Epoch 6/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 50ms/step - accuracy: 0.4102 - loss: 1.8691 - val_accuracy: 0.2740 - val_loss: 2.0237 - learning_rate: 9.0484e-06\n",
      "Epoch 7/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 53ms/step - accuracy: 0.5121 - loss: 1.7234 - val_accuracy: 0.3200 - val_loss: 1.9449 - learning_rate: 8.1873e-06\n",
      "Epoch 8/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 54ms/step - accuracy: 0.5217 - loss: 1.7243 - val_accuracy: 0.3460 - val_loss: 1.8791 - learning_rate: 7.4082e-06\n",
      "Epoch 9/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 49ms/step - accuracy: 0.5466 - loss: 1.6401 - val_accuracy: 0.3680 - val_loss: 1.8269 - learning_rate: 6.7032e-06\n",
      "Epoch 10/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 50ms/step - accuracy: 0.5264 - loss: 1.6181 - val_accuracy: 0.3860 - val_loss: 1.7811 - learning_rate: 6.0653e-06\n",
      "Fold 2\n",
      "Epoch 1/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 60ms/step - accuracy: 0.4565 - loss: 1.7730 - val_accuracy: 0.5640 - val_loss: 1.4883 - learning_rate: 6.0653e-06\n",
      "Epoch 2/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 51ms/step - accuracy: 0.4351 - loss: 1.7127 - val_accuracy: 0.5800 - val_loss: 1.4494 - learning_rate: 6.0653e-06\n",
      "Epoch 3/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 53ms/step - accuracy: 0.4404 - loss: 1.6501 - val_accuracy: 0.6060 - val_loss: 1.4120 - learning_rate: 6.0653e-06\n",
      "Epoch 4/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 49ms/step - accuracy: 0.5333 - loss: 1.6445 - val_accuracy: 0.6160 - val_loss: 1.3803 - learning_rate: 6.0653e-06\n",
      "Epoch 5/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 49ms/step - accuracy: 0.5509 - loss: 1.5857 - val_accuracy: 0.6200 - val_loss: 1.3473 - learning_rate: 6.0653e-06\n",
      "Epoch 6/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 51ms/step - accuracy: 0.4899 - loss: 1.6165 - val_accuracy: 0.6200 - val_loss: 1.3208 - learning_rate: 5.4881e-06\n",
      "Epoch 7/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 52ms/step - accuracy: 0.5586 - loss: 1.5157 - val_accuracy: 0.6280 - val_loss: 1.2980 - learning_rate: 4.9659e-06\n",
      "Epoch 8/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 51ms/step - accuracy: 0.5788 - loss: 1.4776 - val_accuracy: 0.6360 - val_loss: 1.2771 - learning_rate: 4.4933e-06\n",
      "Epoch 9/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 50ms/step - accuracy: 0.5965 - loss: 1.4387 - val_accuracy: 0.6400 - val_loss: 1.2605 - learning_rate: 4.0657e-06\n",
      "Epoch 10/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 50ms/step - accuracy: 0.6060 - loss: 1.4112 - val_accuracy: 0.6460 - val_loss: 1.2450 - learning_rate: 3.6788e-06\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-09-02 19:50:22.550057: I external/local_xla/xla/stream_executor/cuda/cuda_asm_compiler.cc:393] ptxas warning : Registers are spilled to local memory in function 'gemm_fusion_dot_1169', 8 bytes spill stores, 8 bytes spill loads\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Test Accuracy: 0.5699999928474426\n",
      "WARNING:tensorflow:5 out of the last 15 calls to <function TensorFlowTrainer.make_predict_function.<locals>.one_step_on_data_distributed at 0x7fdf2a131300> triggered tf.function retracing. Tracing is expensive and the excessive number of tracings could be due to (1) creating @tf.function repeatedly in a loop, (2) passing tensors with different shapes, (3) passing Python objects instead of tensors. For (1), please define your @tf.function outside of the loop. For (2), @tf.function has reduce_retracing=True option that can avoid unnecessary retracing. For (3), please refer to https://www.tensorflow.org/guide/function#controlling_retracing and https://www.tensorflow.org/api_docs/python/tf/function for  more details.\n",
      "\u001b[1m7/7\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 199ms/step\n",
      "Precision: 0.5869781144781144\n",
      "Recall: 0.57\n",
      "F1-Score: 0.5679252766473905\n",
      "ROC-AUC: 0.8973518931972926\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1000x700 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# cell6: Transfer learning using MobileNetV2 with K-Fold Cross-Validation for robust evaluation\n",
    "\n",
    "import numpy as np  # Importing numpy for numerical operations\n",
    "import tensorflow as tf  # Importing TensorFlow for deep learning\n",
    "from tensorflow.keras.applications import MobileNetV2  # Importing MobileNetV2 pre-trained model\n",
    "from tensorflow.keras.models import Model  # Importing Model class for creating a model\n",
    "from tensorflow.keras.layers import Dense, GlobalAveragePooling2D  # Importing layers for building the model\n",
    "from tensorflow.keras.callbacks import LearningRateScheduler, EarlyStopping  # Importing callbacks for learning rate scheduling and early stopping\n",
    "from tensorflow.keras.preprocessing.image import ImageDataGenerator  # Importing ImageDataGenerator for data augmentation\n",
    "from sklearn.model_selection import KFold  # Importing KFold for cross-validation\n",
    "from sklearn.metrics import confusion_matrix, precision_score, recall_score, f1_score, roc_auc_score  # Importing metrics for evaluation\n",
    "import matplotlib.pyplot as plt  # Importing matplotlib for plotting\n",
    "import seaborn as sns  # Importing seaborn for enhanced visualizations\n",
    "\n",
    "# Load and preprocess data\n",
    "(x_train, y_train), (x_test, y_test) = tf.keras.datasets.cifar10.load_data()  # Loading the CIFAR-10 dataset\n",
    "x_train, x_test = x_train / 255.0, x_test / 255.0  # Normalizing the pixel values to the range [0, 1]\n",
    "y_train = y_train.flatten()  # Flattening the training labels\n",
    "y_test = y_test.flatten()  # Flattening the test labels\n",
    "\n",
    "# Use a smaller subset of the data for testing\n",
    "x_train_small = x_train[:1000]  # Using the first 1000 training images\n",
    "y_train_small = y_train[:1000]  # Using the first 1000 training labels\n",
    "x_test_small = x_test[:200]  # Using the first 200 test images\n",
    "y_test_small = y_test[:200]  # Using the first 200 test labels\n",
    "\n",
    "# Resize images to (224, 224) for MobileNetV2\n",
    "x_train_small = np.array([tf.image.resize(img, [224, 224]) for img in x_train_small])  # Resizing training images to (224, 224)\n",
    "x_test_small = np.array([tf.image.resize(img, [224, 224]) for img in x_test_small])  # Resizing test images to (224, 224)\n",
    "\n",
    "# Data Augmentation\n",
    "datagen = ImageDataGenerator(  # Creating an ImageDataGenerator for data augmentation\n",
    "    width_shift_range=0.1,  # Allowing horizontal shifts\n",
    "    height_shift_range=0.1,  # Allowing vertical shifts\n",
    "    horizontal_flip=True  # Allowing horizontal flips\n",
    ")\n",
    "datagen.fit(x_train_small)  # Fitting the data generator on the training data\n",
    "\n",
    "# Load pre-trained MobileNetV2 model and add custom layers\n",
    "base_model = MobileNetV2(weights='imagenet', include_top=False, input_shape=(224, 224, 3))  # Loading MobileNetV2 with pre-trained weights and excluding the top classification layer\n",
    "x = base_model.output  # Getting the output of the base model\n",
    "x = GlobalAveragePooling2D()(x)  # Adding a global average pooling layer to reduce spatial dimensions\n",
    "x = Dense(128, activation='relu')(x)  # Adding a dense layer with 128 units and ReLU activation\n",
    "predictions = Dense(10, activation='softmax')(x)  # Adding the output layer with softmax activation\n",
    "\n",
    "model = Model(inputs=base_model.input, outputs=predictions)  # Creating the model\n",
    "\n",
    "# Freeze the base model layers\n",
    "for layer in base_model.layers:  # Iterating over the layers of the base model\n",
    "    layer.trainable = False  # Freezing the layers to prevent further training\n",
    "\n",
    "# Unfreeze the top layers\n",
    "for layer in base_model.layers[-4:]:  # Iterating over the last 4 layers of the base model\n",
    "    layer.trainable = True  # Unfreezing the layers for fine-tuning\n",
    "\n",
    "# Compile the model\n",
    "model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=1e-5), loss='sparse_categorical_crossentropy', metrics=['accuracy'])  # Compiling the model with Adam optimizer and sparse categorical cross-entropy loss\n",
    "\n",
    "# Learning Rate Scheduler\n",
    "def lr_scheduler(epoch, lr):  # Defining a learning rate schedule function\n",
    "    if epoch < 5:  # If epoch number is less than 5\n",
    "        return lr  # Return the current learning rate\n",
    "    else:  # Otherwise\n",
    "        return float(lr * tf.math.exp(-0.1))  # Reduce the learning rate by a factor of exp(-0.1)\n",
    "\n",
    "# Early Stopping\n",
    "early_stopping = EarlyStopping(monitor='val_loss', patience=5, restore_best_weights=True)  # Creating an EarlyStopping callback\n",
    "\n",
    "# K-Fold Cross-Validation\n",
    "kfold = KFold(n_splits=2, shuffle=True, random_state=42)  # Creating a KFold cross-validator with 2 splits\n",
    "fold_no = 1  # Initializing the fold number\n",
    "history = []  # Initializing a list to store the training history\n",
    "\n",
    "for train, val in kfold.split(x_train_small, y_train_small):  # Iterating over the folds\n",
    "    print(f'Fold {fold_no}')  # Printing the current fold number\n",
    "    \n",
    "    # Train the model with reduced batch size and fewer epochs\n",
    "    history.append(model.fit(  # Training the model\n",
    "        datagen.flow(x_train_small[train], y_train_small[train], batch_size=16),  # Using the data generator to generate augmented data with a batch size of 16\n",
    "        epochs=10,  # Number of epochs to train\n",
    "        validation_data=(x_train_small[val], y_train_small[val]),  # Validation data\n",
    "        callbacks=[LearningRateScheduler(lr_scheduler), early_stopping],  # Using the defined callbacks\n",
    "        verbose=1  # Verbosity level\n",
    "    ))\n",
    "    \n",
    "    fold_no += 1  # Incrementing the fold number\n",
    "\n",
    "# Evaluate model on test set\n",
    "test_loss, test_accuracy = model.evaluate(x_test_small, y_test_small, verbose=0)  # Evaluating the model on the test set\n",
    "print(f'Test Accuracy: {test_accuracy}')  # Printing the test accuracy\n",
    "\n",
    "# Calculate additional metrics\n",
    "y_pred = model.predict(x_test_small)  # Predicting on the test set\n",
    "y_pred_classes = np.argmax(y_pred, axis=1)  # Getting the predicted classes\n",
    "\n",
    "conf_matrix = confusion_matrix(y_test_small, y_pred_classes)  # Calculating the confusion matrix\n",
    "precision = precision_score(y_test_small, y_pred_classes, average='weighted')  # Calculating the precision\n",
    "recall = recall_score(y_test_small, y_pred_classes, average='weighted')  # Calculating the recall\n",
    "f1 = f1_score(y_test_small, y_pred_classes, average='weighted')  # Calculating the F1-score\n",
    "\n",
    "# Ensure y_pred are probabilities\n",
    "y_pred_prob = tf.nn.softmax(y_pred, axis=1).numpy()  # Converting the predictions to probabilities\n",
    "roc_auc = roc_auc_score(y_test_small, y_pred_prob, multi_class='ovr')  # Calculating the ROC-AUC score\n",
    "\n",
    "print(f'Precision: {precision}')  # Printing the precision\n",
    "print(f'Recall: {recall}')  # Printing the recall\n",
    "print(f'F1-Score: {f1}')  # Printing the F1-score\n",
    "print(f'ROC-AUC: {roc_auc}')  # Printing the ROC-AUC score\n",
    "\n",
    "# Visualize confusion matrix\n",
    "plt.figure(figsize=(10, 7))  # Creating a figure for plotting\n",
    "sns.heatmap(conf_matrix, annot=True, fmt='d', cmap='Blues', xticklabels=range(10), yticklabels=range(10))  # Plotting the confusion matrix\n",
    "plt.xlabel('Predicted')  # Setting the x-axis label\n",
    "plt.ylabel('True')  # Setting the y-axis label\n",
    "plt.title('Confusion Matrix')  # Setting the title\n",
    "plt.show()  # Displaying the plot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fold 1\n",
      "Epoch 1/10\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/matt/CIFAR10_CNN_Project/venv/lib/python3.12/site-packages/keras/src/trainers/data_adapters/py_dataset_adapter.py:122: UserWarning: Your `PyDataset` class should call `super().__init__(**kwargs)` in its constructor. `**kwargs` can include `workers`, `use_multiprocessing`, `max_queue_size`. Do not pass these arguments to `fit()`, as they will be ignored.\n",
      "  self._warn_if_super_not_called()\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m10s\u001b[0m 192ms/step - accuracy: 0.0931 - loss: 2.7021 - val_accuracy: 0.1340 - val_loss: 2.5615 - learning_rate: 1.0000e-05\n",
      "Epoch 2/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 53ms/step - accuracy: 0.1439 - loss: 2.5181 - val_accuracy: 0.1540 - val_loss: 2.4210 - learning_rate: 1.0000e-05\n",
      "Epoch 3/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 49ms/step - accuracy: 0.1506 - loss: 2.4717 - val_accuracy: 0.1900 - val_loss: 2.3042 - learning_rate: 1.0000e-05\n",
      "Epoch 4/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 48ms/step - accuracy: 0.2460 - loss: 2.2579 - val_accuracy: 0.2040 - val_loss: 2.2046 - learning_rate: 1.0000e-05\n",
      "Epoch 5/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 53ms/step - accuracy: 0.2121 - loss: 2.2450 - val_accuracy: 0.2480 - val_loss: 2.1167 - learning_rate: 1.0000e-05\n",
      "Epoch 6/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 47ms/step - accuracy: 0.2318 - loss: 2.2102 - val_accuracy: 0.3020 - val_loss: 2.0467 - learning_rate: 9.0484e-06\n",
      "Epoch 7/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 51ms/step - accuracy: 0.2346 - loss: 2.1206 - val_accuracy: 0.3280 - val_loss: 1.9880 - learning_rate: 8.1873e-06\n",
      "Epoch 8/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 50ms/step - accuracy: 0.2632 - loss: 2.0971 - val_accuracy: 0.3500 - val_loss: 1.9384 - learning_rate: 7.4082e-06\n",
      "Epoch 9/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 48ms/step - accuracy: 0.2819 - loss: 1.9725 - val_accuracy: 0.3740 - val_loss: 1.8951 - learning_rate: 6.7032e-06\n",
      "Epoch 10/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 48ms/step - accuracy: 0.2586 - loss: 2.0023 - val_accuracy: 0.3900 - val_loss: 1.8604 - learning_rate: 6.0653e-06\n",
      "Fold 2\n",
      "Epoch 1/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m6s\u001b[0m 167ms/step - accuracy: 0.2174 - loss: 2.1031 - val_accuracy: 0.4180 - val_loss: 1.6804 - learning_rate: 6.0653e-06\n",
      "Epoch 2/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 48ms/step - accuracy: 0.2871 - loss: 2.0593 - val_accuracy: 0.4360 - val_loss: 1.6499 - learning_rate: 6.0653e-06\n",
      "Epoch 3/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 49ms/step - accuracy: 0.3179 - loss: 2.0153 - val_accuracy: 0.4480 - val_loss: 1.6222 - learning_rate: 6.0653e-06\n",
      "Epoch 4/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 51ms/step - accuracy: 0.3141 - loss: 1.9805 - val_accuracy: 0.4740 - val_loss: 1.5954 - learning_rate: 6.0653e-06\n",
      "Epoch 5/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 52ms/step - accuracy: 0.3432 - loss: 1.9064 - val_accuracy: 0.4960 - val_loss: 1.5688 - learning_rate: 6.0653e-06\n",
      "Epoch 6/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 49ms/step - accuracy: 0.3245 - loss: 1.9275 - val_accuracy: 0.5120 - val_loss: 1.5470 - learning_rate: 5.4881e-06\n",
      "Epoch 7/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 47ms/step - accuracy: 0.3348 - loss: 1.8857 - val_accuracy: 0.5220 - val_loss: 1.5271 - learning_rate: 4.9659e-06\n",
      "Epoch 8/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 50ms/step - accuracy: 0.3637 - loss: 1.8141 - val_accuracy: 0.5240 - val_loss: 1.5098 - learning_rate: 4.4933e-06\n",
      "Epoch 9/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 50ms/step - accuracy: 0.3117 - loss: 1.8963 - val_accuracy: 0.5260 - val_loss: 1.4950 - learning_rate: 4.0657e-06\n",
      "Epoch 10/10\n",
      "\u001b[1m32/32\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 48ms/step - accuracy: 0.3362 - loss: 1.8767 - val_accuracy: 0.5320 - val_loss: 1.4824 - learning_rate: 3.6788e-06\n",
      "Test Accuracy: 0.4650000035762787\n",
      "\u001b[1m7/7\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 182ms/step\n",
      "Precision: 0.4773119939972881\n",
      "Recall: 0.465\n",
      "F1-Score: 0.44683877724308074\n",
      "ROC-AUC: 0.8682563313454572\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1000x700 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# cell7: Transfer learning using MobileNetV2 with dropout regularization and K-Fold Cross-Validation for robust evaluation\n",
    "\n",
    "import numpy as np  # Importing numpy for numerical operations\n",
    "import tensorflow as tf  # Importing TensorFlow for deep learning\n",
    "from tensorflow.keras.applications import MobileNetV2  # Importing MobileNetV2 pre-trained model\n",
    "from tensorflow.keras.models import Model  # Importing Model class for creating a model\n",
    "from tensorflow.keras.layers import Dense, GlobalAveragePooling2D, Dropout  # Importing layers for building the model\n",
    "from tensorflow.keras.callbacks import LearningRateScheduler, EarlyStopping  # Importing callbacks for learning rate scheduling and early stopping\n",
    "from tensorflow.keras.preprocessing.image import ImageDataGenerator  # Importing ImageDataGenerator for data augmentation\n",
    "from sklearn.model_selection import KFold  # Importing KFold for cross-validation\n",
    "from sklearn.metrics import confusion_matrix, precision_score, recall_score, f1_score, roc_auc_score  # Importing metrics for evaluation\n",
    "import matplotlib.pyplot as plt  # Importing matplotlib for plotting\n",
    "import seaborn as sns  # Importing seaborn for enhanced visualizations\n",
    "\n",
    "# Load and preprocess data\n",
    "(x_train, y_train), (x_test, y_test) = tf.keras.datasets.cifar10.load_data()  # Loading the CIFAR-10 dataset\n",
    "x_train, x_test = x_train / 255.0, x_test / 255.0  # Normalizing the pixel values to the range [0, 1]\n",
    "y_train = y_train.flatten()  # Flattening the training labels\n",
    "y_test = y_test.flatten()  # Flattening the test labels\n",
    "\n",
    "# Use a smaller subset of the data for testing\n",
    "x_train_small = x_train[:1000]  # Using the first 1000 training images\n",
    "y_train_small = y_train[:1000]  # Using the first 1000 training labels\n",
    "x_test_small = x_test[:200]  # Using the first 200 test images\n",
    "y_test_small = y_test[:200]  # Using the first 200 test labels\n",
    "\n",
    "# Resize images to (224, 224) for MobileNetV2\n",
    "x_train_small = np.array([tf.image.resize(img, [224, 224]) for img in x_train_small])  # Resizing training images to (224, 224)\n",
    "x_test_small = np.array([tf.image.resize(img, [224, 224]) for img in x_test_small])  # Resizing test images to (224, 224)\n",
    "\n",
    "# Data Augmentation\n",
    "datagen = ImageDataGenerator(  # Creating an ImageDataGenerator for data augmentation\n",
    "    width_shift_range=0.1,  # Allowing horizontal shifts\n",
    "    height_shift_range=0.1,  # Allowing vertical shifts\n",
    "    horizontal_flip=True  # Allowing horizontal flips\n",
    ")\n",
    "datagen.fit(x_train_small)  # Fitting the data generator on the training data\n",
    "\n",
    "# Load pre-trained MobileNetV2 model and add custom layers\n",
    "base_model = MobileNetV2(weights='imagenet', include_top=False, input_shape=(224, 224, 3))  # Loading MobileNetV2 with pre-trained weights and excluding the top classification layer\n",
    "x = base_model.output  # Getting the output of the base model\n",
    "x = GlobalAveragePooling2D()(x)  # Adding a global average pooling layer to reduce spatial dimensions\n",
    "x = Dense(128, activation='relu')(x)  # Adding a dense layer with 128 units and ReLU activation\n",
    "x = Dropout(0.5)(x)  # Adding dropout for regularization\n",
    "predictions = Dense(10, activation='softmax')(x)  # Adding the output layer with softmax activation\n",
    "\n",
    "model = Model(inputs=base_model.input, outputs=predictions)  # Creating the model\n",
    "\n",
    "# Freeze the base model layers\n",
    "for layer in base_model.layers:  # Iterating over the layers of the base model\n",
    "    layer.trainable = False  # Freezing the layers to prevent further training\n",
    "\n",
    "# Unfreeze the top layers\n",
    "for layer in base_model.layers[-4:]:  # Iterating over the last 4 layers of the base model\n",
    "    layer.trainable = True  # Unfreezing the layers for fine-tuning\n",
    "\n",
    "# Compile the model\n",
    "model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=1e-5), loss='sparse_categorical_crossentropy', metrics=['accuracy'])  # Compiling the model with Adam optimizer and sparse categorical cross-entropy loss\n",
    "\n",
    "# Learning Rate Scheduler\n",
    "def lr_scheduler(epoch, lr):  # Defining a learning rate schedule function\n",
    "    if epoch < 5:  # If epoch number is less than 5\n",
    "        return lr  # Return the current learning rate\n",
    "    else:  # Otherwise\n",
    "        return float(lr * tf.math.exp(-0.1))  # Reduce the learning rate by a factor of exp(-0.1)\n",
    "\n",
    "# Early Stopping\n",
    "early_stopping = EarlyStopping(monitor='val_loss', patience=5, restore_best_weights=True)  # Creating an EarlyStopping callback\n",
    "\n",
    "# K-Fold Cross-Validation\n",
    "kfold = KFold(n_splits=2, shuffle=True, random_state=42)  # Creating a KFold cross-validator with 2 splits\n",
    "fold_no = 1  # Initializing the fold number\n",
    "history = []  # Initializing a list to store the training history\n",
    "\n",
    "for train, val in kfold.split(x_train_small, y_train_small):  # Iterating over the folds\n",
    "    print(f'Fold {fold_no}')  # Printing the current fold number\n",
    "    \n",
    "    # Train the model with reduced batch size and fewer epochs\n",
    "    history.append(model.fit(  # Training the model\n",
    "        datagen.flow(x_train_small[train], y_train_small[train], batch_size=16),  # Using the data generator to generate augmented data with a batch size of 16\n",
    "        epochs=10,  # Number of epochs to train\n",
    "        validation_data=(x_train_small[val], y_train_small[val]),  # Validation data\n",
    "        callbacks=[LearningRateScheduler(lr_scheduler), early_stopping],  # Using the defined callbacks\n",
    "        verbose=1  # Verbosity level\n",
    "    ))\n",
    "    \n",
    "    fold_no += 1  # Incrementing the fold number\n",
    "\n",
    "# Evaluate model on test set\n",
    "test_loss, test_accuracy = model.evaluate(x_test_small, y_test_small, verbose=0)  # Evaluating the model on the test set\n",
    "print(f'Test Accuracy: {test_accuracy}')  # Printing the test accuracy\n",
    "\n",
    "# Calculate additional metrics\n",
    "y_pred = model.predict(x_test_small)  # Predicting on the test set\n",
    "y_pred_classes = np.argmax(y_pred, axis=1)  # Getting the predicted classes\n",
    "\n",
    "conf_matrix = confusion_matrix(y_test_small, y_pred_classes)  # Calculating the confusion matrix\n",
    "precision = precision_score(y_test_small, y_pred_classes, average='weighted')  # Calculating the precision\n",
    "recall = recall_score(y_test_small, y_pred_classes, average='weighted')  # Calculating the recall\n",
    "f1 = f1_score(y_test_small, y_pred_classes, average='weighted')  # Calculating the F1-score\n",
    "\n",
    "# Ensure y_pred are probabilities\n",
    "y_pred_prob = tf.nn.softmax(y_pred, axis=1).numpy()  # Converting the predictions to probabilities\n",
    "roc_auc = roc_auc_score(y_test_small, y_pred_prob, multi_class='ovr')  # Calculating the ROC-AUC score\n",
    "\n",
    "print(f'Precision: {precision}')  # Printing the precision\n",
    "print(f'Recall: {recall}')  # Printing the recall\n",
    "print(f'F1-Score: {f1}')  # Printing the F1-score\n",
    "print(f'ROC-AUC: {roc_auc}')  # Printing the ROC-AUC score\n",
    "\n",
    "# Visualize confusion matrix\n",
    "plt.figure(figsize=(10, 7))  # Creating a figure for plotting\n",
    "sns.heatmap(conf_matrix, annot=True, fmt='d', cmap='Blues', xticklabels=range(10), yticklabels=range(10))  # Plotting the confusion matrix\n",
    "plt.xlabel('Predicted')  # Setting the x-axis label\n",
    "plt.ylabel('True')  # Setting the y-axis label\n",
    "plt.title('Confusion Matrix')  # Setting the title\n",
    "plt.show()  # Displaying the plot"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "venv",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.12.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}