slides.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "slideshow": {
     "slide_type": "skip"
    }
   },
   "outputs": [],
   "source": [
    "from IPython.core.display import HTML\n",
    "HTML(\"\"\"\n",
    "<style>\n",
    "\n",
    "div.text_cell_render h1 { /* Main titles bigger, centered */\n",
    "text-align:center;\n",
    "}\n",
    "\n",
    "</style>\n",
    "\"\"\")\n",
    "\n",
    "from matplotlib import pyplot as plt\n",
    "%matplotlib inline\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "source": [
    "![](images/tensorflow.jpg)\n",
    "\n",
    "![](https://www.android.com/static/2016/img/share/andy-lg.png)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "source": [
    "![](images/android-tensorflow-app-structure_2.png)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "source": [
    "![](images/android_tensorflow_classifier_results.jpg)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "source": [
    "![](images/example.png)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "source": [
    "![](images/mnist.jpeg)\n",
    "\n",
    "Классическая задача распознавания рукописных цифр на основе набора данных [MNIST](http://yann.lecun.com/exdb/mnist/)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x131534978>"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1315d1e48>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from model import data\n",
    "x_train, y_train, x_test, y_test = data.load_data()\n",
    "plt.imshow(x_train[0].reshape((28,28)), cmap='gray')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {
    "slideshow": {
     "slide_type": "subslide"
    }
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "from keras.datasets import mnist\n",
    "from keras.utils import to_categorical\n",
    "\n",
    "\n",
    "def load_data(flatten=True):\n",
    "    \"\"\"\n",
    "    Load MNIST dataset, do train-test split, convert data to the necessary format\n",
    "    :return:\n",
    "    \"\"\"\n",
    "    if flatten:\n",
    "        reshape = _flatten\n",
    "    else:\n",
    "        reshape = _square\n",
    "\n",
    "    (x_train, y_train), (x_test, y_test) = mnist.load_data()\n",
    "    x_train = reshape(x_train)\n",
    "    x_test = reshape(x_test)\n",
    "    x_train = x_train.astype('float32')\n",
    "    x_test = x_test.astype('float32')\n",
    "    x_train /= 255\n",
    "    x_test /= 255\n",
    "    y_train = to_categorical(y_train, 10)\n",
    "    y_test = to_categorical(y_test, 10)\n",
    "    return x_train, y_train, x_test, y_test\n",
    "\n",
    "\n",
    "def _flatten(x: np.ndarray) -> np.ndarray:\n",
    "    return x.reshape(x.shape[0], 28 * 28)\n",
    "\n",
    "\n",
    "def _square(x: np.ndarray) -> np.ndarray:\n",
    "    return x.reshape(x.shape[0], 28, 28, 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "source": [
    "# Что используем\n",
    "\n",
    "- [TensorFlow backend](https://www.tensorflow.org/)\n",
    "- [Keras](https://keras.io/)\n",
    "- Android SDK > 23\n",
    "- [org.tensorflow:tensorflow-android:1.3.0](https://www.tensorflow.org/mobile/android_build)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {
    "slideshow": {
     "slide_type": "subslide"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "numpy\r\n",
      "tensorflow\r\n",
      "keras\r\n",
      "h5py"
     ]
    }
   ],
   "source": [
    "!cat model/requirements.txt"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "source": [
    "# 1. Neural Network"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "source": [
    "# Простейшая однослойная нейросеть"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "fragment"
    }
   },
   "source": [
    "![](images/one_layer.png)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {
    "slideshow": {
     "slide_type": "-"
    }
   },
   "outputs": [],
   "source": [
    "import keras\n",
    "keras.backend.clear_session()\n",
    "tf.reset_default_graph()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_1 (Dense)              (None, 10)                7850      \n",
      "=================================================================\n",
      "Total params: 7,850\n",
      "Trainable params: 7,850\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "from keras import metrics, Sequential\n",
    "from keras.layers import Dense\n",
    "\n",
    "model = Sequential(layers=[\n",
    "    Dense(10, activation='softmax', input_shape=(784,))\n",
    "])\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "outputs": [],
   "source": [
    "from keras.layers import K\n",
    "from keras.optimizers import SGD\n",
    "\n",
    "model.compile(loss=K.categorical_crossentropy,\n",
    "              optimizer=SGD(),\n",
    "              metrics=[metrics.categorical_accuracy])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 60000 samples, validate on 10000 samples\n",
      "Epoch 1/10\n",
      "60000/60000 [==============================] - 2s 30us/step - loss: 1.2668 - categorical_accuracy: 0.7063 - val_loss: 0.8042 - val_categorical_accuracy: 0.8336\n",
      "Epoch 2/10\n",
      "60000/60000 [==============================] - 1s 22us/step - loss: 0.7117 - categorical_accuracy: 0.8414 - val_loss: 0.6047 - val_categorical_accuracy: 0.8602\n",
      "Epoch 3/10\n",
      "60000/60000 [==============================] - 2s 31us/step - loss: 0.5852 - categorical_accuracy: 0.8595 - val_loss: 0.5249 - val_categorical_accuracy: 0.8755\n",
      "Epoch 4/10\n",
      "60000/60000 [==============================] - 2s 28us/step - loss: 0.5244 - categorical_accuracy: 0.8687 - val_loss: 0.4794 - val_categorical_accuracy: 0.8805\n",
      "Epoch 5/10\n",
      "60000/60000 [==============================] - 2s 26us/step - loss: 0.4871 - categorical_accuracy: 0.8751 - val_loss: 0.4500 - val_categorical_accuracy: 0.8858\n",
      "Epoch 6/10\n",
      "60000/60000 [==============================] - 2s 26us/step - loss: 0.4615 - categorical_accuracy: 0.8802 - val_loss: 0.4286 - val_categorical_accuracy: 0.8881\n",
      "Epoch 7/10\n",
      "60000/60000 [==============================] - 1s 25us/step - loss: 0.4424 - categorical_accuracy: 0.8835 - val_loss: 0.4129 - val_categorical_accuracy: 0.8917\n",
      "Epoch 8/10\n",
      "60000/60000 [==============================] - 2s 25us/step - loss: 0.4276 - categorical_accuracy: 0.8862 - val_loss: 0.3999 - val_categorical_accuracy: 0.8948\n",
      "Epoch 9/10\n",
      "60000/60000 [==============================] - 1s 24us/step - loss: 0.4157 - categorical_accuracy: 0.8884 - val_loss: 0.3897 - val_categorical_accuracy: 0.8976\n",
      "Epoch 10/10\n",
      "60000/60000 [==============================] - 2s 26us/step - loss: 0.4057 - categorical_accuracy: 0.8909 - val_loss: 0.3811 - val_categorical_accuracy: 0.8991\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x13330e668>"
      ]
     },
     "execution_count": 47,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(x_train, y_train,\n",
    "          batch_size=128,\n",
    "          epochs=10,\n",
    "          validation_data=(x_test, y_test))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "source": [
    "# Многослойная нейросеть"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "fragment"
    }
   },
   "source": [
    "![](images/deep_neural_network.png)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_1 (Dense)              (None, 200)               157000    \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 10)                2010      \n",
      "=================================================================\n",
      "Total params: 159,010\n",
      "Trainable params: 159,010\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "keras.backend.clear_session()\n",
    "tf.reset_default_graph()\n",
    "\n",
    "model = Sequential(layers=[\n",
    "    Dense(200, activation='relu', input_shape=(784,)),\n",
    "    Dense(10, activation='softmax')\n",
    "])\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 60000 samples, validate on 10000 samples\n",
      "Epoch 1/10\n",
      "60000/60000 [==============================] - 2s 40us/step - loss: 1.1825 - categorical_accuracy: 0.7202 - val_loss: 0.6388 - val_categorical_accuracy: 0.8595\n",
      "Epoch 2/10\n",
      "60000/60000 [==============================] - 2s 41us/step - loss: 0.5502 - categorical_accuracy: 0.8672 - val_loss: 0.4509 - val_categorical_accuracy: 0.8903\n",
      "Epoch 3/10\n",
      "60000/60000 [==============================] - 3s 47us/step - loss: 0.4361 - categorical_accuracy: 0.8865 - val_loss: 0.3853 - val_categorical_accuracy: 0.9010\n",
      "Epoch 4/10\n",
      "60000/60000 [==============================] - 2s 42us/step - loss: 0.3857 - categorical_accuracy: 0.8961 - val_loss: 0.3507 - val_categorical_accuracy: 0.9072\n",
      "Epoch 5/10\n",
      "60000/60000 [==============================] - 2s 41us/step - loss: 0.3557 - categorical_accuracy: 0.9029 - val_loss: 0.3284 - val_categorical_accuracy: 0.9117\n",
      "Epoch 6/10\n",
      "60000/60000 [==============================] - 2s 39us/step - loss: 0.3348 - categorical_accuracy: 0.9078 - val_loss: 0.3120 - val_categorical_accuracy: 0.9138\n",
      "Epoch 7/10\n",
      "60000/60000 [==============================] - 2s 41us/step - loss: 0.3187 - categorical_accuracy: 0.9116 - val_loss: 0.2987 - val_categorical_accuracy: 0.9188\n",
      "Epoch 8/10\n",
      "60000/60000 [==============================] - 3s 46us/step - loss: 0.3055 - categorical_accuracy: 0.9149 - val_loss: 0.2883 - val_categorical_accuracy: 0.9211\n",
      "Epoch 9/10\n",
      "60000/60000 [==============================] - 3s 49us/step - loss: 0.2944 - categorical_accuracy: 0.9176 - val_loss: 0.2793 - val_categorical_accuracy: 0.9237\n",
      "Epoch 10/10\n",
      "60000/60000 [==============================] - 3s 52us/step - loss: 0.2846 - categorical_accuracy: 0.9200 - val_loss: 0.2707 - val_categorical_accuracy: 0.9257\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x13190a470>"
      ]
     },
     "execution_count": 50,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from keras.callbacks import TensorBoard\n",
    "\n",
    "model.compile(loss=K.categorical_crossentropy,\n",
    "              optimizer=SGD(),\n",
    "              metrics=[metrics.categorical_accuracy])\n",
    "\n",
    "model.fit(x_train, y_train,\n",
    "          batch_size=128,\n",
    "          epochs=10,\n",
    "          validation_data=(x_test, y_test),\n",
    "          callbacks=[TensorBoard(log_dir='/tmp/tensorflow/dnn_jup')])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "source": [
    "# Tensorboard"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "subslide"
    }
   },
   "source": [
    "![](images/tensorboard_graph.png)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "subslide"
    }
   },
   "source": [
    "![](images/tensorboard_scalars.png)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "source": [
    "# Экспорт модели"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "fragment"
    }
   },
   "source": [
    "- pbtxt\n",
    "- chkp\n",
    "- pb\n",
    "- **optimal pb** for android\n",
    "- ..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Restoring parameters from out/dnn.chkp\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Restoring parameters from out/dnn.chkp\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Froze 4 variables.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Froze 4 variables.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Converted 4 variables to const ops.\n",
      "15 ops in the final graph.\n",
      "graph saved!\n"
     ]
    }
   ],
   "source": [
    "from model import exporter\n",
    "\n",
    "exporter.export_model(tf.train.Saver(), \n",
    "                      input_names=['dense_1_input'], \n",
    "                      output_name='dense_2/Softmax', \n",
    "                      model_name=\"dnn_jup\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {
    "slideshow": {
     "slide_type": "subslide"
    }
   },
   "outputs": [],
   "source": [
    "def export_model(saver: tf.train.Saver, input_names: list, output_name: str, model_name: str):\n",
    "    \"\"\"\n",
    "    You can find node names by using debugger: just connect it right after model is created and look for nodes in the inspec\n",
    "    :param saver:\n",
    "    :param input_names:\n",
    "    :param output_name:\n",
    "    :param model_name:\n",
    "    :return:\n",
    "    \"\"\"\n",
    "    os.makedirs(\"./out\", exist_ok=True)\n",
    "    tf.train.write_graph(K.get_session().graph_def, 'out',\n",
    "                         model_name + '_graph.pbtxt')\n",
    "\n",
    "    saver.save(K.get_session(), 'out/' + model_name + '.chkp')\n",
    "\n",
    "    # pbtxt is human readable representation of the graph\n",
    "    freeze_graph.freeze_graph('out/' + model_name + '_graph.pbtxt', None,\n",
    "                              False, 'out/' + model_name + '.chkp', output_name,\n",
    "                              \"save/restore_all\", \"save/Const:0\",\n",
    "                              'out/frozen_' + model_name + '.pb', True, \"\")\n",
    "\n",
    "    input_graph_def = tf.GraphDef()\n",
    "    with tf.gfile.Open('out/frozen_' + model_name + '.pb', \"rb\") as f:\n",
    "        input_graph_def.ParseFromString(f.read())\n",
    "\n",
    "    # optimization of the graph so we can use it in the android app\n",
    "    output_graph_def = optimize_for_inference_lib.optimize_for_inference(\n",
    "        input_graph_def, input_names, [output_name],\n",
    "        tf.float32.as_datatype_enum)\n",
    "\n",
    "    # This is archived optimal graph in the protobuf format we'll use in our android App.\n",
    "    with tf.gfile.FastGFile('out/opt_' + model_name + '.pb', \"wb\") as f:\n",
    "        f.write(output_graph_def.SerializeToString())\n",
    "\n",
    "    print(\"graph saved!\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "checkpoint                   dnn_graph.pbtxt\r\n",
      "dnn.chkp.data-00000-of-00001 frozen_dnn.pb\r\n",
      "dnn.chkp.index               opt_dnn.pb\r\n",
      "dnn.chkp.meta\r\n"
     ]
    }
   ],
   "source": [
    "!ls out/"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "source": [
    "# Свёрточная нейосеть"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "fragment"
    }
   },
   "source": [
    "![](images/common_cnn_design.jpg)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "source": [
    "![](images/conv_layers.png)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras.layers import Conv2D, MaxPooling2D\n",
    "from keras.layers import Flatten"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "outputs": [],
   "source": [
    "keras.backend.clear_session()\n",
    "tf.reset_default_graph()\n",
    "\n",
    "model = Sequential()\n",
    "model.add(Conv2D(filters=16, kernel_size=5, strides=1,\n",
    "                 padding='same', activation='relu',\n",
    "                 input_shape=[28, 28, 1]))\n",
    "# 28*28*64\n",
    "model.add(MaxPooling2D(pool_size=2, strides=2, padding='same'))\n",
    "# 14*14*64\n",
    "\n",
    "model.add(Conv2D(filters=32, kernel_size=4, strides=1,\n",
    "                 padding='same', activation='relu'))\n",
    "# 14*14*128\n",
    "model.add(MaxPooling2D(pool_size=2, strides=2, padding='same'))\n",
    "# 7*7*128\n",
    "\n",
    "model.add(Conv2D(filters=64, kernel_size=3, strides=1,\n",
    "                 padding='same', activation='relu'))\n",
    "# 7*7*256\n",
    "model.add(MaxPooling2D(pool_size=2, strides=2, padding='same'))\n",
    "# 4*4*256\n",
    "\n",
    "model.add(Flatten())\n",
    "model.add(Dense(64 * 4, activation='relu'))\n",
    "model.add(Dense(10, activation='softmax'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_train, y_train, x_test, y_test = data.load_data(flatten=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [],
   "source": [
    "!rm -rf /tmp/tensorflow/conv_jup\n",
    "from keras.optimizers import Adam\n",
    "\n",
    "model.compile(loss=K.categorical_crossentropy,\n",
    "              optimizer=Adam(),\n",
    "              metrics=[metrics.categorical_accuracy])\n",
    "\n",
    "model.fit(x_train, y_train,\n",
    "          batch_size=128,\n",
    "          epochs=1,\n",
    "          validation_data=(x_test, y_test),\n",
    "          callbacks=[TensorBoard(log_dir='/tmp/tensorflow/conv_jup')])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "exporter.export_model(tf.train.Saver(), \n",
    "                      input_names=['conv2d_1_input'], \n",
    "                      output_name='dense_2/Softmax', \n",
    "                      model_name=\"conv_jup\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "source": [
    "# 2. Android"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "fragment"
    }
   },
   "source": [
    "![](images/example.png)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "slideshow": {
     "slide_type": "slide"
    }
   },
   "source": [
    "# 3. Интеграция"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "```groovy\n",
    "implementation 'org.tensorflow:tensorflow-android:1.3.0'\n",
    "```"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "```kotlin\n",
    "import org.senior_sigan.digits.views.DrawView\n",
    "import org.senior_sigan.digits.ml.models.DigitsClassifierFlatten // for DNN\n",
    "import org.senior_sigan.digits.ml.models.DigitsClassifierSquare // for Conv\n",
    "```"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "```kotlin\n",
    "// somewhere in another thread load the model\n",
    "val clf = DigitsClassifierFlatten(\n",
    "    assetManager=assets, \n",
    "    modelPath=\"dnn.pb\", \n",
    "    inputSize=28, \n",
    "    inputName=\"dense_1_input\", \n",
    "    outputName=\"dense_2/Softmax\", \n",
    "    name=\"dnn\")\n",
    "\n",
    "// somewhere on UI thread\n",
    "val drawView = find<DrawView>(R.id.draw)\n",
    "\n",
    "// somewhere on another thread\n",
    "val pred = clf.predict(drawView.pixelData)\n",
    "\n",
    "Log.i(\"Dnn\", \"${pred.label} ${pred.proba}\")\n",
    "\n",
    "```"
   ]
  }
 ],
 "metadata": {
  "celltoolbar": "Slideshow",
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.6.3"
  },
  "livereveal": {
   "scroll": true
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}