monitor_training.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "49608b32",
   "metadata": {},
   "source": [
    "This notebook allows to look at the data produced when models are trained with the main.py program.\n",
    "\n",
    "It allows to monitor in real time the training of the models when we run main.py.\n",
    "\n",
    "It is also useful because it allows to look at the training data of any trained models when the training is done."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "04ae9db8-9551-4429-b70c-fc203365a421",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import numpy as np\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import matplotlib.pyplot as plt\n",
    "from torch.utils.data import DataLoader, TensorDataset, Dataset\n",
    "from torch.optim.lr_scheduler import ReduceLROnPlateau\n",
    "from copy import deepcopy\n",
    "from pathlib import Path\n",
    "import torch.nn.functional as F\n",
    "import random"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "51b2454f-fdf9-4974-bd7f-c9cabe41d49f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Using device: cpu\n"
     ]
    }
   ],
   "source": [
    "device =  \"cuda\" if torch.cuda.is_available() else \"cpu\"\n",
    "print(f\"Using device: {device}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "f53fab39",
   "metadata": {},
   "outputs": [],
   "source": [
    "run = 1 # select here the run\n",
    "H = 34 # select here the hidden layer size of the model you want to see "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "087893f4-be1e-4fc9-8018-439beb8dbffa",
   "metadata": {},
   "outputs": [],
   "source": [
    "K = 10 # Number of classes\n",
    "\n",
    "class FCNeuralNetwork(nn.Module):\n",
    "    def __init__(self):\n",
    "        super().__init__()\n",
    "        self.Matrix1 = nn.Linear(28**2, H)\n",
    "        self.Matrix2 = nn.Linear(H, K)\n",
    "        self.R = nn.ReLU()\n",
    "    def forward(self,x):\n",
    "        x = x.view(-1, 28**2)\n",
    "        x = self.R(self.Matrix1(x))\n",
    "        x = self.Matrix2(x)\n",
    "        return x.squeeze()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "5f985846-fc0f-4db0-b014-37fb99283b61",
   "metadata": {},
   "outputs": [],
   "source": [
    "#loading training data\n",
    "model = FCNeuralNetwork().to(device)\n",
    "model.load_state_dict(torch.load(f'runs/run{1}/H{H}/model.pt'))\n",
    "train_squared_loss_numpy=np.load(f'runs/run{1}/H{H}/train_squared_loss_numpy.npy')\n",
    "val_squared_loss_numpy=np.load(f'runs/run{1}/H{H}/val_squared_loss_numpy.npy')\n",
    "train_zero_one_loss_numpy=np.load(f'runs/run{1}/H{H}/train_zero_one_loss_numpy.npy')\n",
    "val_zero_one_loss_numpy=np.load(f'runs/run{1}/H{H}/val_zero_one_loss_numpy.npy')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "a120df99-291e-4922-be70-2afeffaa1403",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch number:25\n"
     ]
    }
   ],
   "source": [
    "print(f\"epoch number:{train_squared_loss_numpy.shape[0]}\")  #here we look at how many epochs have been done"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "279f4428-3547-4ffb-819e-ceceed9b318a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x258e0eef070>"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# here we plot the evolution of the train and val squared loss at each epoch.\n",
    "\n",
    "plt.figure()\n",
    "plt.plot(train_squared_loss_numpy, label='train loss')\n",
    "plt.plot(val_squared_loss_numpy, label='validation loss')\n",
    "plt.xlabel('nb epoch')\n",
    "#plt.yscale(\"log\")\n",
    "plt.ylabel('loss')\n",
    "plt.legend()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "33244d21",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x258e11bdab0>"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# here we plot the evolution of the train and val zero one loss at each epoch.\n",
    "\n",
    "plt.figure()\n",
    "plt.plot(train_zero_one_loss_numpy, label='train loss')\n",
    "plt.plot(val_zero_one_loss_numpy, label='validation loss')\n",
    "plt.xlabel('nb epoch')\n",
    "#plt.yscale(\"log\")\n",
    "plt.ylabel('loss')\n",
    "plt.legend()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "d47e5408",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.045882936507936505\n"
     ]
    }
   ],
   "source": [
    "print(train_zero_one_loss_numpy[-1]) #last train zero one loss"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "6a3f8306",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.09329927898943424\n"
     ]
    }
   ],
   "source": [
    "print(val_zero_one_loss_numpy[-1]) #last test zero one loss"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "019424bd",
   "metadata": {},
   "source": [
    "In what follows we compute the train and test accuracy of the model obtained in the last epoch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "1ce7556c",
   "metadata": {},
   "outputs": [],
   "source": [
    "class ImageDataset(Dataset):\n",
    "    def __init__(self, x, y):\n",
    "        self.x, self.y = x, y\n",
    "        self.x = self.x / 255.  # We normalize the image amplitude between 0 and 1\n",
    "        self.y = F.one_hot(self.y, num_classes=K).to(torch.float) # One hot encoding\n",
    "    def __len__(self): \n",
    "        return self.x.shape[0]\n",
    "    def __getitem__(self, ix): \n",
    "        return self.x[ix], self.y[ix]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "fbd38844",
   "metadata": {},
   "outputs": [],
   "source": [
    "x_train_subsampled=torch.load('dataset/x_train_subsampled.pt')\n",
    "y_train_subsampled=torch.load('dataset/y_train_subsampled.pt')\n",
    "\n",
    "x_test_subsampled=torch.load('dataset/x_test_subsampled.pt')\n",
    "y_test_subsampled=torch.load('dataset/y_test_subsampled.pt')\n",
    "\n",
    "train_dataset = ImageDataset(x_train_subsampled, y_train_subsampled)\n",
    "test_dataset = ImageDataset(x_test_subsampled, y_test_subsampled)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "a2ff3231",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "accuracy = 0.953\n"
     ]
    }
   ],
   "source": [
    "# accuracy on the train dataset of the last model\n",
    "n=0\n",
    "for i in range(len(train_dataset)):\n",
    "    if (torch.argmax(model(train_dataset[i][0].to(device))).item() == y_train_subsampled[i]):\n",
    "        n+=1\n",
    "\n",
    "print(f'accuracy = {n/len(train_dataset)}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "a0ce1b51",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "accuracy = 0.907\n"
     ]
    }
   ],
   "source": [
    "# accuracy on the test dataset of the last model\n",
    "n=0\n",
    "for i in range(len(test_dataset)):\n",
    "    if (torch.argmax(model(test_dataset[i][0].to(device))).item() == y_test_subsampled[i]):\n",
    "        n+=1\n",
    "\n",
    "print(f'accuracy = {n/len(test_dataset)}')"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "fil_project",
   "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": "3.10.16"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}