Intro to torch.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Введение в torch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "\n",
    "from sklearn.metrics import classification_report\n",
    "\n",
    "import torch\n",
    "from torch.utils.data import DataLoader\n",
    "from torchvision.datasets import MNIST\n",
    "from torchvision import transforms\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from tqdm import tqdm\n",
    "\n",
    "from sklearn.metrics import accuracy_score\n",
    "\n",
    "import time\n",
    "\n",
    "import torch"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Датасет\n",
    "Мы его просто загрузим и здесь пока будет какая-то торчовая магия, которую мы разберем на следующем занятии"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Некоторые магические трансформации, о которых пока не стоит заморачиваться\n",
    "1. Перевод картинки в тензор\n",
    "1. Нормализация картинки"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "transform = transforms.Compose([\n",
    "    transforms.ToTensor(),\n",
    "    transforms.Normalize((0.1307,), (0.3081,))\n",
    "])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Загружаем датасет\n",
    "Конкретно эта часть кода понадобится нам один раз за курс и будет только в этой домашке"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/bobazooba/opt/anaconda3/lib/python3.8/site-packages/torchvision/datasets/mnist.py:498: UserWarning: The given NumPy array is not writeable, and PyTorch does not support non-writeable tensors. This means you can write to the underlying (supposedly non-writeable) NumPy array using the tensor. You may want to copy the array to protect its data or make it writeable before converting it to a tensor. This type of warning will be suppressed for the rest of this program. (Triggered internally at  ../torch/csrc/utils/tensor_numpy.cpp:180.)\n",
      "  return torch.from_numpy(parsed.astype(m[2], copy=False)).view(*s)\n"
     ]
    }
   ],
   "source": [
    "train_dataset = MNIST('.', train=True, download=True, transform=transform)\n",
    "test_dataset = MNIST('.', train=False, transform=transform)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Переводим его в два класса\n",
    "Это нужно для упрощения задачи до бинарной классификации. В этой домашке мы будем отличать две цифры: 1 и 7."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_dataset.data = train_dataset.data[(train_dataset.targets == 3) | (train_dataset.targets == 7)]\n",
    "train_dataset.targets = train_dataset.targets[(train_dataset.targets == 3) | (train_dataset.targets == 7)]\n",
    "\n",
    "test_dataset.data = test_dataset.data[(test_dataset.targets == 3) | (test_dataset.targets == 7)]\n",
    "test_dataset.targets = test_dataset.targets[(test_dataset.targets == 3) | (test_dataset.targets == 7)]\n",
    "\n",
    "train_dataset.targets = (train_dataset.targets == 7).long()\n",
    "test_dataset.targets = (test_dataset.targets == 7).long()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Оборачиваем в итератор данных\n",
    "Далее в курсе мы поговорим о том, что это такое более подробно, а дальше в этой тетрадке будет несколько полезных деталей.  \n",
    "Пока что будет достаточно понимать, что мы обернули наши данные в некоторую абстракцию, итерация по которой дает нам батчи данных из нашего датасета.  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_loader = DataLoader(train_dataset, batch_size=128, shuffle=True)\n",
    "test_loader = DataLoader(test_dataset, batch_size=128, shuffle=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Порисуем данные\n",
    "\n",
    "- классу 0 соответствует цифра 3\n",
    "- классу 1 соответствует цифра 7  \n",
    "\n",
    "Выбрал эти цифры, потому что на них хуже всего метрики"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x504 with 25 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "X, y = next(iter(train_loader))\n",
    "X = X.numpy()\n",
    "y = y.numpy()\n",
    "\n",
    "plt.figure(figsize=(6, 7))\n",
    "for i in range(25):\n",
    "    plt.subplot(5, 5, i+1)\n",
    "    plt.imshow(X[i].reshape(28, 28), cmap=plt.cm.Greys_r)\n",
    "    plt.title(y[i])\n",
    "    plt.axis('off')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Мы можем определить нашу модель как последовательность уже существующих слоев, как в керасе"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Sequential(\n",
       "  (linear_1): Linear(in_features=784, out_features=256, bias=True)\n",
       "  (relu): ReLU()\n",
       "  (linear_2): Linear(in_features=256, out_features=1, bias=True)\n",
       "  (sigmoid): Sigmoid()\n",
       ")"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model = torch.nn.Sequential()\n",
    "\n",
    "model.add_module(name='linear_1', module=torch.nn.Linear(in_features=784, out_features=256))\n",
    "model.add_module(name='relu', module=torch.nn.ReLU())\n",
    "model.add_module(name='linear_2', module=torch.nn.Linear(in_features=256, out_features=1))\n",
    "model.add_module(name='sigmoid', module=torch.nn.Sigmoid())\n",
    "\n",
    "model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Либо так"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Sequential(\n",
       "  (0): Linear(in_features=784, out_features=256, bias=True)\n",
       "  (1): ReLU()\n",
       "  (2): Linear(in_features=256, out_features=1, bias=True)\n",
       "  (3): Sigmoid()\n",
       ")"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model = torch.nn.Sequential(\n",
    "    torch.nn.Linear(in_features=784, out_features=256),\n",
    "    torch.nn.ReLU(),\n",
    "    torch.nn.Linear(in_features=256, out_features=1),\n",
    "    torch.nn.Sigmoid()\n",
    ")\n",
    "\n",
    "model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## А вот для чего нужно было писать в домашке классы\n",
    "## Для определения своих моделей/слоев/чего вам захочется нам нужно сделать такую структуру"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "# наследуемся от класса torch.nn.Module\n",
    "class MLP(torch.nn.Module):\n",
    "    \n",
    "    # определяем метод __init__\n",
    "    # он нужен для того, чтобы мы могли определить нужные нам слои, веса, сохраняемые операции\n",
    "    # \n",
    "    # передаем ссылку на самого себя первым параметром\n",
    "    # \n",
    "    # определяем любые другие передаваемые аргументы\n",
    "    # могут быть чем угодно: числа, словари, другие слои, другие модели и вообще все что угодно\n",
    "    # \n",
    "    def __init__(self, in_features, inner_features, out_features):\n",
    "        \n",
    "        # Вызываем __init__ родителя - torch.nn.Module\n",
    "        super().__init__()\n",
    "        \n",
    "        # определяем слои и все что захотим сохранять/использовать\n",
    "        self.linear_1 = torch.nn.Linear(in_features=in_features, out_features=inner_features)\n",
    "        self.non_linear_function = torch.nn.ReLU()\n",
    "        self.linear_2 = torch.nn.Linear(in_features=inner_features, out_features=out_features)\n",
    "        self.sigmoid = torch.nn.Sigmoid()\n",
    "        \n",
    "    # нам нужно обязательно реализовать метод forward\n",
    "    # это то, что нашей модели нужно делать, как оперировать с входными данными, как рассчитывать\n",
    "    # метод backward нам не нужно реализовывать, потому что в торче есть автоматическое дифференцирование\n",
    "    # главное, чтобы наши операции были дифференцируемы\n",
    "    # некоторые подробности можете узнать здесь:\n",
    "    # https://pytorch.org/tutorials/beginner/blitz/autograd_tutorial.html#sphx-glr-beginner-blitz-autograd-tutorial-py\n",
    "    # также список доступных дифференцируемых операций можете посмотреть здесь:\n",
    "    # https://pytorch.org/docs/stable/tensors.html\n",
    "    def forward(self, x):\n",
    "        \n",
    "        # наша привычная схема\n",
    "        x = self.linear_1(x)\n",
    "        x = self.non_linear_function(x)\n",
    "        x = self.linear_2(x)\n",
    "        x = self.sigmoid(x)\n",
    "        \n",
    "        return x\n",
    "    \n",
    "    # мы также можем реализовать любые другие полезные нам методы внутри класса"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "# определяем нашу модель\n",
    "model = MLP(in_features=784, inner_features=256, out_features=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "MLP(\n",
       "  (linear_1): Linear(in_features=784, out_features=256, bias=True)\n",
       "  (non_linear_function): ReLU()\n",
       "  (linear_2): Linear(in_features=256, out_features=1, bias=True)\n",
       "  (sigmoid): Sigmoid()\n",
       ")"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([256, 784])"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# можем посмотреть веса\n",
    "model.linear_1.weight.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "# еще нам нужно задать функцию потерь \n",
    "criterion = torch.nn.BCELoss()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "# и оптимизатор, куда нужно передать параметры модели\n",
    "# параметры модели можно взять методом класса .parameters()\n",
    "optimizer = torch.optim.SGD(params=model.parameters(), lr=0.01)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[Parameter containing:\n",
       " tensor([[ 0.0145, -0.0216,  0.0109,  ...,  0.0296, -0.0126,  0.0122],\n",
       "         [-0.0332,  0.0219, -0.0287,  ..., -0.0222, -0.0106,  0.0294],\n",
       "         [ 0.0231, -0.0329,  0.0002,  ..., -0.0297,  0.0315,  0.0324],\n",
       "         ...,\n",
       "         [-0.0315,  0.0279,  0.0336,  ...,  0.0259,  0.0098, -0.0082],\n",
       "         [-0.0065, -0.0318, -0.0222,  ..., -0.0339, -0.0328,  0.0025],\n",
       "         [-0.0213,  0.0315,  0.0204,  ..., -0.0069, -0.0279,  0.0024]],\n",
       "        requires_grad=True),\n",
       " Parameter containing:\n",
       " tensor([ 1.6150e-02,  3.1855e-02,  2.4850e-02, -3.2633e-03, -3.4994e-02,\n",
       "         -2.3166e-02,  6.7476e-03, -2.5939e-02,  7.2354e-03, -1.7827e-02,\n",
       "          1.8702e-02, -3.5713e-02,  7.0094e-03, -7.5102e-04, -2.9308e-02,\n",
       "          2.9857e-02, -1.7982e-02, -1.1262e-02, -4.8488e-03, -3.2002e-02,\n",
       "         -4.3904e-03,  2.7694e-02,  9.3303e-03,  3.6537e-03, -2.7550e-02,\n",
       "         -2.6065e-02,  5.4980e-03, -3.4676e-02,  1.7932e-02, -4.6961e-03,\n",
       "         -3.5074e-02,  5.5640e-03, -3.1190e-02,  2.9811e-02, -2.9557e-02,\n",
       "          1.9794e-02, -4.5555e-03, -1.8863e-02,  2.8937e-02, -1.0108e-02,\n",
       "          2.5664e-02,  3.3291e-02,  1.9646e-02, -2.0673e-02,  2.1811e-02,\n",
       "         -1.8602e-03,  9.3866e-03,  2.8490e-02, -2.7089e-02, -2.9950e-02,\n",
       "          2.3301e-02, -3.9543e-03, -3.2078e-02, -3.0307e-03, -2.8556e-02,\n",
       "         -4.4173e-04, -4.8686e-03,  6.8638e-03, -3.6634e-03,  3.2985e-03,\n",
       "         -2.0125e-02, -3.5693e-02, -1.4479e-02, -5.1323e-03, -2.6234e-02,\n",
       "          3.2719e-03,  3.2817e-02,  2.1662e-02, -1.8972e-02,  1.1291e-02,\n",
       "         -2.6424e-02,  2.1556e-02,  1.5840e-02, -1.0425e-02,  2.8213e-02,\n",
       "         -1.1970e-02,  5.9786e-03, -3.6766e-03,  6.4106e-03, -3.1063e-02,\n",
       "          2.8623e-02, -1.3420e-03, -2.0564e-02, -1.4442e-02, -2.7510e-02,\n",
       "         -3.2885e-02, -2.9989e-02, -3.2727e-02, -3.1918e-02, -3.0402e-02,\n",
       "          1.9849e-02,  3.3256e-02, -1.5595e-02,  6.6048e-03,  6.2064e-03,\n",
       "          1.9235e-02, -2.9991e-02,  3.2875e-02, -1.7372e-02, -3.5349e-02,\n",
       "         -3.4629e-02,  2.2126e-02, -8.6249e-03,  1.4562e-02, -6.7740e-04,\n",
       "          2.5794e-02,  3.7048e-03, -3.2596e-02,  1.3915e-02, -1.1715e-02,\n",
       "         -3.9322e-03,  6.5050e-03, -1.5668e-02, -1.8716e-02,  6.4094e-04,\n",
       "          1.6700e-03, -2.6930e-02, -2.4882e-02,  6.2382e-03, -3.0177e-02,\n",
       "          1.9701e-02,  3.1048e-02,  2.1795e-02, -8.4510e-03, -2.8919e-02,\n",
       "          2.2354e-02,  1.3592e-02, -5.8129e-03, -1.9294e-02, -1.4938e-02,\n",
       "          3.4737e-02,  3.0992e-02,  3.5540e-02, -3.3773e-02, -6.4348e-03,\n",
       "         -5.0458e-03,  1.6810e-02, -2.2908e-02, -1.3414e-02,  3.8981e-03,\n",
       "          3.5705e-02,  4.9605e-03, -1.5747e-02, -1.0475e-02,  3.0857e-02,\n",
       "         -2.3472e-02, -2.3577e-02,  1.4085e-03,  2.3064e-02, -3.6837e-03,\n",
       "         -1.9990e-02, -3.2267e-02,  1.1608e-02,  1.4953e-02, -2.4817e-02,\n",
       "         -7.1364e-03, -2.2838e-02,  1.0104e-02,  1.1877e-03,  2.9841e-02,\n",
       "         -3.8541e-03,  1.4323e-02,  1.4859e-02,  3.9251e-03,  1.1797e-02,\n",
       "          3.0017e-02,  3.3300e-02, -2.5319e-02, -2.5612e-02,  2.8467e-02,\n",
       "          2.1883e-02,  2.0194e-04, -1.4777e-02,  8.9419e-03, -3.3203e-02,\n",
       "         -1.9284e-02,  2.2711e-02, -1.7602e-02,  2.0660e-02, -2.9638e-02,\n",
       "         -3.2831e-02, -1.6581e-03, -7.7807e-03, -3.1499e-02,  1.2600e-02,\n",
       "          1.4551e-02,  1.5492e-02,  2.6987e-02, -3.6120e-03, -5.6982e-03,\n",
       "         -3.1098e-02,  1.7997e-02, -3.5101e-02, -6.9424e-03, -8.2902e-03,\n",
       "         -3.5381e-02, -8.0808e-03,  2.1239e-02, -5.5546e-03,  2.2667e-03,\n",
       "          2.1703e-02,  1.5427e-02,  2.7491e-02, -2.4073e-02, -3.0597e-02,\n",
       "         -1.9428e-03, -5.7113e-03,  7.6508e-03,  2.6815e-02,  8.3948e-03,\n",
       "         -1.6590e-02, -1.2836e-02,  2.7297e-02,  1.5148e-02, -1.2625e-02,\n",
       "         -9.1490e-03, -2.0772e-02,  1.6089e-02, -9.5677e-05,  2.5235e-02,\n",
       "          1.8120e-02,  3.1917e-02, -1.6887e-02, -1.4155e-02, -1.3564e-02,\n",
       "         -1.2749e-03, -8.4509e-03, -4.2854e-03, -2.1103e-02, -1.4659e-02,\n",
       "         -2.2571e-03, -9.2850e-03,  3.0450e-03, -2.0091e-02,  2.2248e-02,\n",
       "          3.5055e-02,  2.6632e-02, -1.5921e-02,  2.6656e-02, -1.4350e-02,\n",
       "          1.3331e-02, -3.3068e-02, -6.3729e-03,  2.4788e-02, -5.8843e-03,\n",
       "         -8.9088e-03,  6.2614e-03, -4.3615e-03,  1.2952e-02, -2.2266e-02,\n",
       "         -1.9344e-02,  2.1618e-02, -2.5649e-03,  2.6965e-02,  2.3928e-02,\n",
       "         -3.2617e-02], requires_grad=True),\n",
       " Parameter containing:\n",
       " tensor([[-0.0535, -0.0204, -0.0046, -0.0437, -0.0558, -0.0398, -0.0527,  0.0473,\n",
       "          -0.0397, -0.0089, -0.0618, -0.0464,  0.0143, -0.0133,  0.0211, -0.0230,\n",
       "          -0.0613,  0.0443, -0.0062,  0.0394,  0.0165, -0.0102,  0.0066, -0.0584,\n",
       "           0.0040,  0.0291, -0.0183,  0.0229,  0.0483, -0.0581, -0.0421, -0.0411,\n",
       "           0.0206, -0.0319, -0.0126, -0.0345,  0.0155, -0.0595,  0.0051,  0.0301,\n",
       "          -0.0338,  0.0478, -0.0078,  0.0144,  0.0444,  0.0468, -0.0450,  0.0538,\n",
       "           0.0595,  0.0211, -0.0055, -0.0500,  0.0415, -0.0398,  0.0368,  0.0218,\n",
       "           0.0290,  0.0268,  0.0159,  0.0614, -0.0081, -0.0487,  0.0293,  0.0137,\n",
       "           0.0109, -0.0136,  0.0485, -0.0105, -0.0071,  0.0019,  0.0045,  0.0415,\n",
       "           0.0059, -0.0126,  0.0204, -0.0425, -0.0618, -0.0184, -0.0061,  0.0247,\n",
       "          -0.0381,  0.0486,  0.0562,  0.0571, -0.0085,  0.0143,  0.0532,  0.0047,\n",
       "           0.0013,  0.0497,  0.0224,  0.0147, -0.0336,  0.0453, -0.0372,  0.0029,\n",
       "          -0.0100, -0.0088,  0.0379,  0.0421, -0.0599, -0.0514,  0.0614,  0.0245,\n",
       "          -0.0345,  0.0068,  0.0389, -0.0280, -0.0300,  0.0136,  0.0177,  0.0211,\n",
       "           0.0138,  0.0023,  0.0259, -0.0009, -0.0591,  0.0268,  0.0467,  0.0098,\n",
       "           0.0119,  0.0010, -0.0406,  0.0005,  0.0148,  0.0237, -0.0115, -0.0541,\n",
       "           0.0577,  0.0218,  0.0326, -0.0189, -0.0357,  0.0234, -0.0212,  0.0509,\n",
       "           0.0564,  0.0060,  0.0092,  0.0220, -0.0360,  0.0160,  0.0035,  0.0254,\n",
       "          -0.0277, -0.0390,  0.0223,  0.0406,  0.0254,  0.0240,  0.0317, -0.0362,\n",
       "          -0.0297, -0.0605, -0.0174,  0.0352, -0.0285, -0.0299, -0.0519,  0.0436,\n",
       "          -0.0233,  0.0054,  0.0609, -0.0396, -0.0513,  0.0407,  0.0144,  0.0538,\n",
       "           0.0267, -0.0112,  0.0369, -0.0369,  0.0314,  0.0547, -0.0151, -0.0414,\n",
       "           0.0600,  0.0224, -0.0287, -0.0427,  0.0370, -0.0088, -0.0125, -0.0512,\n",
       "           0.0070,  0.0402,  0.0539, -0.0265, -0.0138, -0.0149, -0.0240,  0.0432,\n",
       "           0.0427,  0.0282, -0.0614, -0.0230,  0.0288, -0.0618,  0.0377, -0.0414,\n",
       "          -0.0249,  0.0466,  0.0571, -0.0078, -0.0119, -0.0400, -0.0029, -0.0057,\n",
       "           0.0120,  0.0178,  0.0361,  0.0495,  0.0266,  0.0017, -0.0398, -0.0163,\n",
       "           0.0602, -0.0322, -0.0277,  0.0040, -0.0203, -0.0243,  0.0178, -0.0114,\n",
       "          -0.0275, -0.0390, -0.0371, -0.0446,  0.0263, -0.0395, -0.0338, -0.0057,\n",
       "          -0.0351, -0.0619,  0.0612,  0.0054,  0.0229,  0.0173,  0.0212, -0.0355,\n",
       "           0.0004, -0.0131, -0.0115,  0.0487,  0.0127,  0.0308, -0.0381,  0.0197,\n",
       "           0.0025,  0.0496,  0.0530,  0.0028, -0.0137, -0.0128,  0.0424,  0.0420]],\n",
       "        requires_grad=True),\n",
       " Parameter containing:\n",
       " tensor([0.0594], requires_grad=True)]"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# можем посмотреть на параметры нашей модели\n",
    "params = list(model.parameters())\n",
    "params"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Обучим нашу модель"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# давайте сохранять значения нашей функции потерь каждую итерацию обучения\n",
    "losses = []\n",
    "\n",
    "# эта штука отдает нам батчи данных\n",
    "# мы поговорим больше про нее позже в курсе\n",
    "for x, y in train_loader:\n",
    "\n",
    "    x = x.view(x.shape[0], -1)\n",
    "    y = y.unsqueeze(1).float()\n",
    "    \n",
    "    # обновим наши градиенты\n",
    "    # будем делать это каждую итерацию обучения\n",
    "    # потому что торч сохраняет в себе все градиенты пока не удалите таким образом рассчитанные градиенты\n",
    "    # \n",
    "    # НЕОБЯЗАТЕЛЬНЫЕ ДЕТАЛИ\n",
    "    # функция сохранения градиентов нужна для аккамуляции градиентов\n",
    "    # например, когда нам не хватает памяти на большой батч\n",
    "    # мы можем несколько раз прогнать forward на разных батчах\n",
    "    # наши градиенты будут сохраняться\n",
    "    # и только потом обновить через несколько итераций forward обновить наши веса\n",
    "    # таким образом у нас получается больше батч\n",
    "    # \n",
    "    optimizer.zero_grad()\n",
    "    \n",
    "    # \n",
    "    # предсказание нашей модели\n",
    "    # здесь торч сам сохраняет необходимую информацию\n",
    "    # чтобы понять эту логику у нас как раз и была часть домашки про собственные классы\n",
    "    # \n",
    "    # торч строит граф вычислений\n",
    "    # то есть строит граф по тем операциям, которые вы определили в методе forward\n",
    "    # \n",
    "    # то есть торч понимает, что вы сначала применили линейное преобразование\n",
    "    # знает, где лежат веса для этого линейного преобразования\n",
    "    # знает, что вы затем применили relu\n",
    "    # затем опять линейное преобразование\n",
    "    # затем сигмоиду\n",
    "    # и вот таким образом торч выстроил граф\n",
    "    # который нужен будет для backpropogation\n",
    "    # \n",
    "    pred = model(x)\n",
    "    \n",
    "    # \n",
    "    # рассчитываем функцию потерь, добавляя в наш граф вычислений этот узел\n",
    "    # \n",
    "    loss = criterion(pred, y.float())\n",
    "    \n",
    "    # \n",
    "    # вызываем метод backward() от нашей функции потерь (это крайний узел нашего графа вычислений)\n",
    "    # этот метод проходит от конца графа в начало, вычисляя необходимые градиенты\n",
    "    # передаем градиенты на следующие узлы\n",
    "    # для весов сохраняет\n",
    "    # \n",
    "    # то есть это почти то же самое, что мы делали в домашке с классами\n",
    "    # \n",
    "    loss.backward()\n",
    "    \n",
    "    # \n",
    "    # теперь мы знаем необходимые градиенты и можем обновить веса с помощью градиентного спуска\n",
    "    # \n",
    "    optimizer.step()\n",
    "    \n",
    "    # \n",
    "    # сохраним текущее значение функции потерь, чтобы потом могли посмотреть на график\n",
    "    # очень важный момент: нам нужно отсоединить значение нашей фунции потерь от графа вычислений\n",
    "    # чтобы граф не строился дальше на каждой итерации обучения\n",
    "    # так мы рано или поздно переполним память\n",
    "    # поэтому вызываем этот метод\n",
    "    # работает для скаляров\n",
    "    # \n",
    "    # для тензоров работает метод .tolist()\n",
    "    # \n",
    "    current_loss_value = loss.item()\n",
    "    losses.append(loss.item())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Одна эпоха прошла"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1008x864 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(14, 12))\n",
    "plt.plot(losses)\n",
    "plt.grid()\n",
    "plt.title('Training process')\n",
    "plt.xlabel('Iterations')\n",
    "plt.ylabel('Loss function');"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Теперь мы можем что-то предсказать"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "for x, y in test_loader:\n",
    "\n",
    "    x = x.view(x.shape[0], -1)\n",
    "    y = y.unsqueeze(1).float()\n",
    "    \n",
    "    # чтобы не ходить по всему тестовому датасету, прекратим ходить после первой итерации\n",
    "    break"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([128, 784])"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([128, 1])"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "MLP(\n",
       "  (linear_1): Linear(in_features=784, out_features=256, bias=True)\n",
       "  (non_linear_function): ReLU()\n",
       "  (linear_2): Linear(in_features=256, out_features=1, bias=True)\n",
       "  (sigmoid): Sigmoid()\n",
       ")"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# переходим в режим предсказания\n",
    "# \n",
    "# позже в курсе мы разберем зачем конкретно это нужно\n",
    "# некоторые детали:\n",
    "# дропаут больше не зануляет часть значений, \n",
    "# батч норм больше не нормирует каждый батч, а использует накопленную статистику при обучении и тд\n",
    "# \n",
    "model.eval()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "# этот контект позволяет не сохранять необходимую информацию для рассчета градиентов\n",
    "# нам сейчас нужен только forward проход\n",
    "# мы не собираемся обновлять веса\n",
    "with torch.no_grad():\n",
    "\n",
    "    pred = model(x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([128, 1])"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "# избавимся от единичной размерности\n",
    "pred = pred.squeeze()\n",
    "y = y.squeeze()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([128])"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "# таким образом из торчовых тензоров мы можем сделать нумпаевские массивы\n",
    "pred = pred.numpy()\n",
    "y = y.numpy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([1., 1., 0., 1., 0., 0., 1., 1., 1., 0., 0., 1., 0., 1., 0., 1., 1.,\n",
       "       0., 1., 1., 1., 1., 0., 0., 0., 1., 1., 0., 1., 1., 1., 1., 1., 0.,\n",
       "       1., 0., 1., 0., 1., 0., 0., 0., 0., 0., 1., 1., 1., 0., 0., 1., 1.,\n",
       "       0., 1., 1., 1., 1., 0., 1., 1., 1., 0., 0., 0., 0., 1., 0., 0., 1.,\n",
       "       0., 0., 1., 0., 1., 0., 0., 1., 0., 0., 1., 1., 1., 1., 0., 1., 0.,\n",
       "       0., 0., 0., 0., 0., 1., 1., 0., 1., 0., 0., 0., 1., 0., 1., 0., 0.,\n",
       "       0., 0., 0., 1., 1., 1., 1., 0., 0., 0., 0., 0., 0., 1., 1., 0., 0.,\n",
       "       1., 0., 1., 1., 0., 1., 0., 0., 1.], dtype=float32)"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "numpy.ndarray"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "type(y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "# бинаризируем наши предсказания\n",
    "pred[pred > 0.5] = 1\n",
    "pred[pred <= 0.5] = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.9609375"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# посчитаем метрику на батче\n",
    "accuracy_score(y, pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "# заново определим нашу модель и выучимся несколько эпох\n",
    "# будут уже новые параметры\n",
    "model = MLP(in_features=784, inner_features=256, out_features=1)\n",
    "\n",
    "criterion = torch.nn.BCELoss()\n",
    "# мы позже поговорим про этот оптимизационный алгоритм\n",
    "# но вообще это тот же SGD, но с некоторыми хаками\n",
    "optimizer = torch.optim.Adam(params=model.parameters())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Типичный цикл обучения с кучей сохранений лоссов"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 1: 100%|██████████| 12396/12396 [00:02<00:00, 4724.60it/s, train_loss=0.0644]\n",
      "Epoch 2:   6%|▌         | 768/12396 [00:00<00:02, 4365.85it/s, train_loss=0.0485]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Losses: train - 0.064, test = 0.042\n",
      "Test: accuracy - 0.985\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 2: 100%|██████████| 12396/12396 [00:02<00:00, 5538.23it/s, train_loss=0.0259]\n",
      "Epoch 3:   7%|▋         | 896/12396 [00:00<00:01, 5846.46it/s, train_loss=0.0245]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Losses: train - 0.026, test = 0.033\n",
      "Test: accuracy - 0.991\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 3: 100%|██████████| 12396/12396 [00:02<00:00, 5881.50it/s, train_loss=0.0149]\n",
      "Epoch 4:   7%|▋         | 896/12396 [00:00<00:02, 5168.42it/s, train_loss=0.0141]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Losses: train - 0.015, test = 0.020\n",
      "Test: accuracy - 0.993\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 4: 100%|██████████| 12396/12396 [00:02<00:00, 5877.34it/s, train_loss=0.00982]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Losses: train - 0.010, test = 0.022\n",
      "Test: accuracy - 0.993\n",
      "Early stopping\n"
     ]
    }
   ],
   "source": [
    "epochs = 10\n",
    "losses = []\n",
    "# нужно для early stopping, то есть трекать лучший лосс на валидации\n",
    "best_test_loss = 10.\n",
    "\n",
    "for n_epoch in range(epochs):\n",
    "    \n",
    "    train_losses = []\n",
    "    test_losses = []\n",
    "    test_preds = []\n",
    "    test_targets = []\n",
    "    \n",
    "    progress_bar = tqdm(total=len(train_loader.dataset), desc='Epoch {}'.format(n_epoch + 1))\n",
    "    \n",
    "    for x, y in train_loader:\n",
    "        \n",
    "        x = x.view(x.shape[0], -1)\n",
    "        y = y.unsqueeze(1).float()\n",
    "        \n",
    "        optimizer.zero_grad()\n",
    "        \n",
    "        pred = model(x)\n",
    "        \n",
    "        loss = criterion(pred, y.float())\n",
    "        \n",
    "        loss.backward()\n",
    "        \n",
    "        optimizer.step()\n",
    "        \n",
    "        train_losses.append(loss.item())\n",
    "        losses.append(loss.item())\n",
    "        \n",
    "        progress_bar.set_postfix(train_loss = np.mean(losses[-100:]))\n",
    "\n",
    "        progress_bar.update(x.shape[0])\n",
    "        \n",
    "    progress_bar.close()\n",
    "    \n",
    "    for x, y in test_loader:\n",
    "        \n",
    "        x = x.view(x.shape[0], -1)\n",
    "        y = y.unsqueeze(1).float()\n",
    "\n",
    "        with torch.no_grad():\n",
    "            \n",
    "            pred = model(x)\n",
    "            \n",
    "        test_preds.append(pred.numpy())\n",
    "        test_targets.append(y.numpy())\n",
    "\n",
    "        loss = criterion(pred, y)\n",
    "\n",
    "        test_losses.append(loss.item())\n",
    "        \n",
    "    mean_test_loss = np.mean(test_losses)\n",
    "        \n",
    "    print('Losses: train - {:.3f}, test = {:.3f}'.format(np.mean(train_losses), mean_test_loss))\n",
    "    \n",
    "    test_preds = np.concatenate(test_preds)\n",
    "    test_preds = test_preds.squeeze()\n",
    "    test_targets = np.concatenate(test_targets).squeeze()\n",
    "    \n",
    "    test_preds[test_preds > 0.5] = 1\n",
    "    test_preds[test_preds <= 0.5] = 0\n",
    "    \n",
    "    accurary = accuracy_score(test_targets, test_preds)\n",
    "\n",
    "    print('Test: accuracy - {:.3f}'.format(accurary))\n",
    "        \n",
    "    # early stopping\n",
    "    # то есть: если наша функция потерь на валидационном датасете стала расти, то прекращаем обучение\n",
    "    # рост фунции потерь на валидационном датасете говорит о переобучении\n",
    "    if mean_test_loss < best_test_loss:\n",
    "        best_test_loss = mean_test_loss\n",
    "    else:\n",
    "        print('Early stopping')\n",
    "        break    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1008x864 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(14, 12))\n",
    "plt.plot(losses)\n",
    "plt.grid()\n",
    "plt.title('Training process')\n",
    "plt.xlabel('Iterations')\n",
    "plt.ylabel('Loss function');"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# One more deeper\n",
    "## Пойдем глубже раскрывать суть лего в глубоком обучении\n",
    "Атомарной частью торча является ```torch.nn.Parameter``` - обучаемый вес.  \n",
    "По сути это такой же тензор, но только мы говорим торчу, что мы хотим, чтобы торч сохранял градиент для этого веса, \n",
    "чтобы обновить его при градиентном шаге оптимизатора\n",
    "\n",
    "```torch.nn.Parameter(data=None, requires_grad=True)```  \n",
    "В дефолтном варианте для тензора мы не требуем сохранять градиент, но можем потребовать и тогда наш тензор станет обучаемым весом  \n",
    "```tensor(data, dtype=None, device=None, requires_grad=False)```"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Parameter containing:\n",
       "tensor([0.7342, 0.8849, 0.3019, 0.1648, 0.4465, 0.8658, 0.5627, 0.7905, 0.1017,\n",
       "        0.9085, 0.3798, 0.4118, 0.5204, 0.2906, 0.2410, 0.9253, 0.6431, 0.4687,\n",
       "        0.8773, 0.2452, 0.8743, 0.8294, 0.9562, 0.2782, 0.2181, 0.4884, 0.2344,\n",
       "        0.5957, 0.1507, 0.0697, 0.3099, 0.6499, 0.9078, 0.4172, 0.4225, 0.8076,\n",
       "        0.1629, 0.2547, 0.8149, 0.0939, 0.7576, 0.6596, 0.2755, 0.3015, 0.7385,\n",
       "        0.4451, 0.7071, 0.8217, 0.1541, 0.7524], requires_grad=True)"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "parampampam = torch.nn.Parameter(torch.rand(50))\n",
    "parampampam"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "x = torch.rand(32, 50)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "result = parampampam - x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[ 0.6301,  0.2763, -0.4110,  ...,  0.5835, -0.1419,  0.6848],\n",
       "        [-0.2533,  0.8030, -0.6033,  ...,  0.3672, -0.6476,  0.2370],\n",
       "        [-0.0782,  0.0849, -0.5517,  ...,  0.3523, -0.1530, -0.2053],\n",
       "        ...,\n",
       "        [ 0.4981,  0.5589, -0.2720,  ...,  0.4601, -0.7393,  0.5975],\n",
       "        [ 0.5658,  0.6773, -0.5167,  ...,  0.4234, -0.4328, -0.1349],\n",
       "        [ 0.7074,  0.6341,  0.2738,  ...,  0.1545, -0.5257,  0.1724]],\n",
       "       grad_fn=<SubBackward0>)"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "result"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Можно заметить, что у тензора result теперь есть атрибут grad_fn\n",
    "Это функция, по которой нужно рассчитывать градиент  \n",
    "И это говорит о том, что теперь у нас есть граф вычислений"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<SubBackward0 at 0x7f84162dd4c0>"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "result.grad_fn"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### И мы можем продолжать наш граф"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "result_end = torch.sigmoid(result)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[0.6525, 0.5686, 0.3987,  ..., 0.6419, 0.4646, 0.6648],\n",
       "        [0.4370, 0.6906, 0.3536,  ..., 0.5908, 0.3435, 0.5590],\n",
       "        [0.4804, 0.5212, 0.3655,  ..., 0.5872, 0.4618, 0.4488],\n",
       "        ...,\n",
       "        [0.6220, 0.6362, 0.4324,  ..., 0.6130, 0.3232, 0.6451],\n",
       "        [0.6378, 0.6631, 0.3736,  ..., 0.6043, 0.3935, 0.4663],\n",
       "        [0.6698, 0.6534, 0.5680,  ..., 0.5386, 0.3715, 0.5430]],\n",
       "       grad_fn=<SigmoidBackward>)"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "result_end"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Для примера того, когда у нас граф не строится\n",
    "Когда нам нечего оптимизировать!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "z = torch.rand(50)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "result_2 = torch.sigmoid(z - x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[0.6608, 0.5240, 0.5487,  ..., 0.4485, 0.5694, 0.5182],\n",
       "        [0.4460, 0.6509, 0.5008,  ..., 0.3958, 0.4437, 0.4073],\n",
       "        [0.4896, 0.4762, 0.5137,  ..., 0.3923, 0.5667, 0.3063],\n",
       "        ...,\n",
       "        [0.6306, 0.5936, 0.5829,  ..., 0.4182, 0.4212, 0.4964],\n",
       "        [0.6462, 0.6218, 0.5224,  ..., 0.4093, 0.4971, 0.3215],\n",
       "        [0.6779, 0.6116, 0.7069,  ..., 0.3463, 0.4739, 0.3918]])"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "result_2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "None\n"
     ]
    }
   ],
   "source": [
    "# нет ссылки на функцию, по которой можем рассчитать градиент\n",
    "print(result_2.grad_fn)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Обучаемые веса\n",
    "## Теперь мы знаем, что можем сами задавать веса\n",
    "Так давайте сделаем свой класс для линейного преобразования"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "class MyLinear(torch.nn.Module):\n",
    "    \n",
    "    def __init__(self, in_features, out_features):\n",
    "        \n",
    "        super().__init__()\n",
    "        \n",
    "        # задаем матрицу весов и смещение\n",
    "        self.weight = torch.nn.Parameter(torch.rand(in_features, out_features) * 0.01)\n",
    "        self.bias = torch.nn.Parameter(torch.zeros(out_features))\n",
    "        \n",
    "    def forward(self, x):\n",
    "        \n",
    "        return torch.matmul(x, self.weight) + self.bias"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Вот таким образом можно написать свои функции и то, как их дифференцировать\n",
    "На практике встречается супер редко"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "class MyReLU(torch.autograd.Function):\n",
    "\n",
    "    @staticmethod\n",
    "    def forward(ctx, x):\n",
    "\n",
    "        ctx.save_for_backward(x)\n",
    "        \n",
    "        return x.clamp(min=0)\n",
    "\n",
    "    @staticmethod\n",
    "    def backward(ctx, grad_output):\n",
    "\n",
    "        x, = ctx.saved_tensors\n",
    "        grad_input = grad_output.clone()\n",
    "        grad_input[x < 0] = 0\n",
    "        \n",
    "        return grad_input"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "class MLP(torch.nn.Module):\n",
    "    \n",
    "    def __init__(self, in_features, inner_features, out_features):\n",
    "        \n",
    "        super().__init__()\n",
    "        \n",
    "        # слои определяются также как обычно\n",
    "        self.linear_1 = MyLinear(in_features=in_features, out_features=inner_features)\n",
    "        # обратите внимание как определяется своя функция\n",
    "        self.non_linear = MyReLU.apply\n",
    "        self.linear_2 = MyLinear(in_features=inner_features, out_features=out_features)\n",
    "        # еще один момент:\n",
    "        # мы сознательно не определяем сигмоиду\n",
    "        # \n",
    "        # Почему?\n",
    "        # В торче для NLL и Binary Cross Entropy \n",
    "        # есть более оптимальные слои, которые используют некоторые оптимизационные трюки\n",
    "        # \n",
    "        \n",
    "    def forward(self, x):\n",
    "        \n",
    "        x = self.linear_1(x)\n",
    "        x = self.non_linear(x)\n",
    "        x = self.linear_2(x)\n",
    "        \n",
    "        return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = MLP(784, 256, 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "MLP(\n",
       "  (linear_1): MyLinear()\n",
       "  (linear_2): MyLinear()\n",
       ")"
      ]
     },
     "execution_count": 50,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "# вот оптимизация для BCE \n",
    "# эта функция потерь объединяет в себе сигмоиду и BCE\n",
    "# и решает задачу более эффективно\n",
    "\n",
    "# \n",
    "# Для NLLLoss есть это CrossEntropy: https://pytorch.org/docs/stable/nn.html#crossentropyloss\n",
    "# \n",
    "# \n",
    "criterion = torch.nn.BCEWithLogitsLoss()\n",
    "\n",
    "optimizer = torch.optim.Adam(params=model.parameters())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "# определим функцию для сигмоиды, чтобы использовать при тестировании\n",
    "def sigmoid(x):\n",
    "    return 1 / (1 + np.exp(-x))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 1: 100%|██████████| 12396/12396 [00:02<00:00, 5914.89it/s, train_loss=0.179]\n",
      "Epoch 2:   7%|▋         | 896/12396 [00:00<00:01, 5751.20it/s, train_loss=0.141]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Losses: train - 0.179, test = 0.060\n",
      "Test: accuracy - 0.980\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 2: 100%|██████████| 12396/12396 [00:02<00:00, 5801.38it/s, train_loss=0.04]  \n",
      "Epoch 3:   7%|▋         | 896/12396 [00:00<00:01, 5936.15it/s, train_loss=0.0407]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Losses: train - 0.040, test = 0.046\n",
      "Test: accuracy - 0.983\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 3: 100%|██████████| 12396/12396 [00:02<00:00, 5823.56it/s, train_loss=0.03]  \n",
      "Epoch 4:   8%|▊         | 1024/12396 [00:00<00:01, 6091.17it/s, train_loss=0.0279]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Losses: train - 0.030, test = 0.038\n",
      "Test: accuracy - 0.987\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 4: 100%|██████████| 12396/12396 [00:02<00:00, 5432.39it/s, train_loss=0.024] \n",
      "Epoch 5:   7%|▋         | 896/12396 [00:00<00:01, 6218.97it/s, train_loss=0.0232]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Losses: train - 0.023, test = 0.036\n",
      "Test: accuracy - 0.989\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 5: 100%|██████████| 12396/12396 [00:02<00:00, 5408.91it/s, train_loss=0.0168]\n",
      "Epoch 6:   8%|▊         | 1024/12396 [00:00<00:01, 6208.11it/s, train_loss=0.0165]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Losses: train - 0.017, test = 0.031\n",
      "Test: accuracy - 0.992\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 6: 100%|██████████| 12396/12396 [00:02<00:00, 5748.93it/s, train_loss=0.0132]\n",
      "Epoch 7:   8%|▊         | 1024/12396 [00:00<00:02, 5662.61it/s, train_loss=0.0134]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Losses: train - 0.013, test = 0.024\n",
      "Test: accuracy - 0.993\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 7: 100%|██████████| 12396/12396 [00:02<00:00, 5498.62it/s, train_loss=0.0098] \n",
      "Epoch 8:   8%|▊         | 1024/12396 [00:00<00:02, 5507.93it/s, train_loss=0.00988]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Losses: train - 0.010, test = 0.023\n",
      "Test: accuracy - 0.993\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Epoch 8: 100%|██████████| 12396/12396 [00:02<00:00, 5279.21it/s, train_loss=0.00829]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Losses: train - 0.008, test = 0.026\n",
      "Test: accuracy - 0.992\n",
      "Early stopped\n"
     ]
    }
   ],
   "source": [
    "epochs = 10\n",
    "losses = []\n",
    "best_test_loss = 10.\n",
    "\n",
    "for n_epoch in range(epochs):\n",
    "    \n",
    "    train_losses = []\n",
    "    test_losses = []\n",
    "    test_preds = []\n",
    "    test_targets = []\n",
    "    \n",
    "    progress_bar = tqdm(total=len(train_loader.dataset), desc='Epoch {}'.format(n_epoch + 1))\n",
    "    \n",
    "    for x, y in train_loader:\n",
    "        \n",
    "        x = x.view(x.shape[0], -1)\n",
    "        y = y.unsqueeze(1).float()\n",
    "        \n",
    "        optimizer.zero_grad()\n",
    "        \n",
    "        pred = model(x)\n",
    "        \n",
    "        loss = criterion(pred, y.float())\n",
    "        \n",
    "        loss.backward()\n",
    "        \n",
    "        optimizer.step()\n",
    "        \n",
    "        train_losses.append(loss.item())\n",
    "        losses.append(loss.item())\n",
    "        \n",
    "        progress_bar.set_postfix(train_loss = np.mean(losses[-100:]))\n",
    "\n",
    "        progress_bar.update(x.shape[0])\n",
    "        \n",
    "    progress_bar.close()\n",
    "    \n",
    "    for x, y in test_loader:\n",
    "        \n",
    "        x = x.view(x.shape[0], -1)\n",
    "        y = y.unsqueeze(1).float()\n",
    "\n",
    "        with torch.no_grad():\n",
    "            \n",
    "            pred = model(x)\n",
    "            \n",
    "        test_preds.append(pred.numpy())\n",
    "        test_targets.append(y.numpy())\n",
    "\n",
    "        loss = criterion(pred, y)\n",
    "\n",
    "        test_losses.append(loss.item())\n",
    "        \n",
    "    print('Losses: train - {:.3f}, test = {:.3f}'.format(np.mean(train_losses), np.mean(test_losses)))\n",
    "    \n",
    "    test_preds = np.concatenate(test_preds)\n",
    "    # Применяем сигмоиду\n",
    "    test_preds = sigmoid(test_preds.squeeze())\n",
    "    test_targets = np.concatenate(test_targets).squeeze()\n",
    "    \n",
    "    test_preds[test_preds > 0.5] = 1\n",
    "    test_preds[test_preds <= 0.5] = 0\n",
    "    \n",
    "    accurary = accuracy_score(test_targets, test_preds)\n",
    "\n",
    "    print('Test: accuracy - {:.3f}'.format(accurary))\n",
    "        \n",
    "    if np.mean(test_losses) < best_test_loss:\n",
    "        best_test_loss = np.mean(test_losses)\n",
    "    else:\n",
    "        print('Early stopped')\n",
    "        break    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "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": "3.8.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}