01_Basic_Pytorch.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "6c77ac61-90e6-4be5-9bfc-b9485998b56a",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\Onsiter\\.conda\\envs\\nlp\\lib\\site-packages\\tqdm\\auto.py:22: 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"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "from sklearn.datasets import make_regression\n",
    "import pandas as pd\n",
    "from torch.autograd import Variable"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 125,
   "id": "33496149-b9cf-49a1-9eab-61a6c911f820",
   "metadata": {},
   "outputs": [],
   "source": [
    "X, y = make_regression(n_features=4, n_informative=2, random_state=0, shuffle=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "ff5aa486-038f-4338-8168-a7aba4379036",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>x1</th>\n",
       "      <th>x2</th>\n",
       "      <th>x3</th>\n",
       "      <th>x4</th>\n",
       "      <th>y</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1.764052</td>\n",
       "      <td>0.400157</td>\n",
       "      <td>0.978738</td>\n",
       "      <td>2.240893</td>\n",
       "      <td>49.822907</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1.867558</td>\n",
       "      <td>-0.977278</td>\n",
       "      <td>0.950088</td>\n",
       "      <td>-0.151357</td>\n",
       "      <td>4.877286</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>-0.103219</td>\n",
       "      <td>0.410599</td>\n",
       "      <td>0.144044</td>\n",
       "      <td>1.454274</td>\n",
       "      <td>11.914875</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.761038</td>\n",
       "      <td>0.121675</td>\n",
       "      <td>0.443863</td>\n",
       "      <td>0.333674</td>\n",
       "      <td>19.753078</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>1.494079</td>\n",
       "      <td>-0.205158</td>\n",
       "      <td>0.313068</td>\n",
       "      <td>-0.854096</td>\n",
       "      <td>23.607000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>95</th>\n",
       "      <td>-1.698106</td>\n",
       "      <td>0.387280</td>\n",
       "      <td>-2.255564</td>\n",
       "      <td>-1.022507</td>\n",
       "      <td>-21.564910</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>96</th>\n",
       "      <td>0.038631</td>\n",
       "      <td>-1.656715</td>\n",
       "      <td>-0.985511</td>\n",
       "      <td>-1.471835</td>\n",
       "      <td>-55.818011</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>97</th>\n",
       "      <td>1.648135</td>\n",
       "      <td>0.164228</td>\n",
       "      <td>0.567290</td>\n",
       "      <td>-0.222675</td>\n",
       "      <td>39.385820</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>98</th>\n",
       "      <td>-0.353432</td>\n",
       "      <td>-1.616474</td>\n",
       "      <td>-0.291837</td>\n",
       "      <td>-0.761492</td>\n",
       "      <td>-62.477272</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>99</th>\n",
       "      <td>0.857924</td>\n",
       "      <td>1.141102</td>\n",
       "      <td>1.466579</td>\n",
       "      <td>0.852552</td>\n",
       "      <td>56.572129</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>100 rows × 5 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "          x1        x2        x3        x4          y\n",
       "0   1.764052  0.400157  0.978738  2.240893  49.822907\n",
       "1   1.867558 -0.977278  0.950088 -0.151357   4.877286\n",
       "2  -0.103219  0.410599  0.144044  1.454274  11.914875\n",
       "3   0.761038  0.121675  0.443863  0.333674  19.753078\n",
       "4   1.494079 -0.205158  0.313068 -0.854096  23.607000\n",
       "..       ...       ...       ...       ...        ...\n",
       "95 -1.698106  0.387280 -2.255564 -1.022507 -21.564910\n",
       "96  0.038631 -1.656715 -0.985511 -1.471835 -55.818011\n",
       "97  1.648135  0.164228  0.567290 -0.222675  39.385820\n",
       "98 -0.353432 -1.616474 -0.291837 -0.761492 -62.477272\n",
       "99  0.857924  1.141102  1.466579  0.852552  56.572129\n",
       "\n",
       "[100 rows x 5 columns]"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_data"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2465b03d-8b6f-4ec9-a697-4158ef75097c",
   "metadata": {},
   "source": [
    "# I. Basic Concept"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "eb79b18e-9c44-4f65-99df-0cc582be1a9d",
   "metadata": {},
   "source": [
    "## 1. Dataset Class"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "08395b7a-7988-4688-a33f-72c7ae409fec",
   "metadata": {},
   "source": [
    "Thông thường Dataset sẽ được kế thừa từ class Dataset.\n",
    "\n",
    "Tuy nhiên không nhất thiết phải kế thừa, chỉ cần tạo class theo kiến trúc như bên dưới"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "2d97e500-3091-4232-8548-162ccd36564f",
   "metadata": {},
   "outputs": [],
   "source": [
    "class CustomDataset:\n",
    "    def __init__(self, data, target):\n",
    "        self.data = data\n",
    "        self.target = target\n",
    "    \n",
    "    def __len__(self):\n",
    "        \"\"\"\n",
    "        Return length of data\n",
    "        \"\"\"\n",
    "        return len(self.data)\n",
    "    \n",
    "    def __getitem__(self, idx):\n",
    "        \"\"\"\n",
    "        Return sample at idx as dictionary and data type is tensor\n",
    "        \"\"\"\n",
    "        current_sample = self.data[idx, :]\n",
    "        current_target = self.target[idx]\n",
    "        return {\n",
    "            \"sample\": torch.tensor(current_sample, dtype=torch.float),\n",
    "            \"target\": torch.tensor(current_target, dtype=torch.long)\n",
    "        }"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "acdcc4f5-9718-40fd-8aee-bef146c49691",
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset = CustomDataset(X, y)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "15b3a3e4-a58a-4cf7-b2c5-16cd8367b1e3",
   "metadata": {},
   "source": [
    "Từng sample riêng lẻ được lấy ra như 1 list thông thường."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "4e54d2ae-72da-472a-a221-307232054a13",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'sample': tensor([1.7641, 0.4002, 0.9787, 2.2409]), 'target': tensor(49)}"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dataset[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "abc05786-e070-400b-b017-826015545541",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([1.7641, 0.4002, 0.9787, 2.2409])"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dataset[0]['sample']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "0a853b88-a873-428d-b2e0-518df788c2d7",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "100"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dataset.__len__()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "61977df9-627d-4865-a255-f8edec900e3c",
   "metadata": {},
   "source": [
    "## 2. Dataloader\n",
    "\n",
    "Trước khi tới ứng dụng Dataloader, cần tìm hiểu về khái niệm Epoch và Batchsize.\n",
    "\n",
    "**Batch Size**\n",
    "- Khi đưa toàn bộ data trong một lần feed foward vào model, thường việc tính toán sẽ rất chậm, có thể là bất khả thi. \n",
    "- Vì vậy ta đưa 1 lượng data (^2, 8-16-32-64 rows) cho 1 lần thực hiện feed foward. Mặc dù điều này có thể gây vấn đề optimize local (not optimize global). \n",
    "\n",
    "Để dễ dàng chia data set thành nhiều batch size một cách ngầu nhiên, ta có thể sử dụng Dataloader.\n",
    "- Tạo dataset với format của pytorch. Dễ dàng tạo ra nhiều batch khác nhau."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "fc2cc797-e3de-4891-8e99-481481980f9a",
   "metadata": {},
   "outputs": [],
   "source": [
    "from torch.utils.data import DataLoader"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "b66f18c8-c18f-4175-8857-42ee4d728202",
   "metadata": {},
   "outputs": [],
   "source": [
    "data_loader = DataLoader(dataset, batch_size=8, shuffle=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3a4e6581-750d-44ef-8a84-3c19b022937a",
   "metadata": {},
   "source": [
    "Với batch size = 8. Tức data chia thành nhiều batch, trong đó 1 batch sẽ chưa 8 rows data\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "88caba71-4448-4e88-a76d-80bc74a2f5ac",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'sample': tensor([[ 1.7641,  0.4002,  0.9787,  2.2409],\n",
      "        [ 1.8676, -0.9773,  0.9501, -0.1514],\n",
      "        [-0.1032,  0.4106,  0.1440,  1.4543],\n",
      "        [ 0.7610,  0.1217,  0.4439,  0.3337],\n",
      "        [ 1.4941, -0.2052,  0.3131, -0.8541],\n",
      "        [-2.5530,  0.6536,  0.8644, -0.7422],\n",
      "        [ 2.2698, -1.4544,  0.0458, -0.1872],\n",
      "        [ 1.5328,  1.4694,  0.1549,  0.3782]]), 'target': tensor([ 49,   4,  11,  19,  23, -29,  -3,  81])}\n"
     ]
    }
   ],
   "source": [
    "for data in data_loader:\n",
    "    print(data)\n",
    "    break"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "26519946-8133-4b7d-bf48-55d272a708fb",
   "metadata": {},
   "source": [
    "## 3. Gradient and Backward\n",
    "\n",
    "Trước khi build model Linear Regression đơn giản với Pytorch, ta cần làm rõ vấn đề sau :\n",
    "- Cách thức Pytoch tính đạo hàm (Gradient)\n",
    "- Khái niệm về backward\n",
    "- Tính toán Loss function\n",
    "- Optimization với loss function, trong đó sử dụng Gradient. \n",
    "\n",
    "\n",
    "https://www.tutorialspoint.com/what-does-backward-do-in-pytorch\n",
    "\n",
    "https://medium.com/@monadsblog/pytorch-backward-function-e5e2b7e60140\n",
    "\n",
    "https://abishekbashyall.medium.com/playing-with-backward-method-in-pytorch-bd34b58745a0\n",
    "\n",
    "- Có thể tính toán đạo hàm tại một node bằng cách set `requires_grad = True.`\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e434ecbf-59aa-45bd-881f-4ffdfa9b2e10",
   "metadata": {},
   "source": [
    "### 3.1 Basic Gradient and Backward"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "36dde3da-b216-4027-9c0d-dfbbe73982dd",
   "metadata": {},
   "source": [
    "Cho function, F = a*b\n",
    "Với \n",
    "- a = 10 \n",
    "- b = 10\n",
    "\n",
    "> Đạo hàm riêng của a : ∂F/∂a = b => ∂F/∂a = 20\n",
    "\n",
    "> Đạo hàm riêng của b : ∂F/∂b = a => ∂F/∂b = 10\n",
    "\n",
    "**Note**\n",
    "- Đạo hàm chỉ được tính sau khi backward được gọi.\n",
    "- Backward là quá trình tính toán đạo hàm từ tensor thực hiện lệnh backward cho tới toàn bộ tensor tại lớp lá (Graph leaves)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "21fd00f9-bc51-4e19-bf1d-078f9a3cc9db",
   "metadata": {},
   "outputs": [],
   "source": [
    "a = torch.tensor([10.], requires_grad=True)\n",
    "b = torch.tensor([20.], requires_grad=True)\n",
    "\n",
    "x = torch.tensor([20.])\n",
    "\n",
    "f = torch.matmul(a,  b) + x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "39ca1d96-2ae3-4d0a-9246-0776f75ae98d",
   "metadata": {},
   "outputs": [],
   "source": [
    "f.backward()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "b9ecf74c-bda2-4255-b8c9-5e5ea0a4e5d9",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Gradient of a :  tensor([20.]) Gradient b :  tensor([10.])\n"
     ]
    }
   ],
   "source": [
    "print(\"Gradient of a : \", a.grad, \"Gradient b : \", b.grad)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4d4cbf55-87b9-45d0-8005-035c9083aeb4",
   "metadata": {},
   "source": [
    "### 3.2 Loss\n",
    "\n"
   ]
  },
  {
   "attachments": {
    "ca95ae4f-d10d-48e0-a904-c72aa5d093de.png": {
     "image/png": "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"
    }
   },
   "cell_type": "markdown",
   "id": "3c380739-ef47-4884-9f86-f8adfa02e1e2",
   "metadata": {},
   "source": [
    "![image.png](attachment:ca95ae4f-d10d-48e0-a904-c72aa5d093de.png)\n",
    "\n",
    "**Backward()** Computes the gradient of current tensor w.r.t. graph leaves.\n",
    "\n",
    "\n",
    "- Gradient chỉ được tính sau khi function `backward()` được gọi\n",
    "- Chỉ những left node mới có thể tính được gradient. Ví dụ tại `W, b hoặc X`\n",
    "\n",
    "**NOTE**:\n",
    "\n",
    "> Không thể run backward cho z. Nếu z là dim 1,1 thì còn tính được, nhưng nếu dim > 1 thì báo lỗi. Nguyên nhân cần tìm hiểu thêm. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "77fd50a1-ea06-4ed0-873f-3241d7789fef",
   "metadata": {},
   "outputs": [],
   "source": [
    "x = torch.randn(10, 8)  # input tensor. Giả sử ta có 10 samples, mỗi sample gồm 8 feature\n",
    "y = torch.zeros(10)    # Target tương ứng với giá trị y thực tế ứng với 10 samples trên.\n",
    "\n",
    "W = torch.randn(8, requires_grad=True) # weights tương ứng với 8 features trên\n",
    "b = torch.randn(1, requires_grad=True) # bias vector\n",
    "\n",
    "z = torch.matmul(x, W) + b # output của model.\n",
    "\n",
    "loss = torch.nn.functional.binary_cross_entropy_with_logits(z, y)  # Loss values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "56e53e05-b25f-4c35-8f7b-1a4e54eb2d51",
   "metadata": {},
   "outputs": [],
   "source": [
    "loss.backward()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "e0b0b781-460e-48e6-a7a3-273b8ea1c62d",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Gradient of weight is  tensor([ 0.0570,  0.0378, -0.2207, -0.0791, -0.0656, -0.1663, -0.2806,  0.1156]) . Gradient of b is  tensor([-1.4276], requires_grad=True)\n"
     ]
    }
   ],
   "source": [
    "print(\"Gradient of weight is \", W.grad, \". Gradient of b is \", b) "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bb90297a-62af-47c7-81f8-72953c4b9153",
   "metadata": {},
   "source": [
    "## 4. Simple Linear Regression Model"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ec278289-0df2-416a-abf7-2cdb1b6027a9",
   "metadata": {},
   "source": [
    "### 4.1 Generate data\n",
    "\n",
    "- Features : Gồm 4 biến x1 ~ x4. dạng float\n",
    "- Target : Numeric. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "id": "557b7624-aad9-41ba-b91d-f3ef4975bf53",
   "metadata": {},
   "outputs": [],
   "source": [
    "features = ['x1', 'x2', 'x3', 'x4']\n",
    "labels = 'y'\n",
    "df_data = pd.DataFrame(X, columns=features)\n",
    "df_data[labels] = y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 188,
   "id": "1595db4a-548d-4016-9375-a4debf5a4955",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>x1</th>\n",
       "      <th>x2</th>\n",
       "      <th>x3</th>\n",
       "      <th>x4</th>\n",
       "      <th>y</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1.764052</td>\n",
       "      <td>0.400157</td>\n",
       "      <td>0.978738</td>\n",
       "      <td>2.240893</td>\n",
       "      <td>49.822907</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1.867558</td>\n",
       "      <td>-0.977278</td>\n",
       "      <td>0.950088</td>\n",
       "      <td>-0.151357</td>\n",
       "      <td>4.877286</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>-0.103219</td>\n",
       "      <td>0.410599</td>\n",
       "      <td>0.144044</td>\n",
       "      <td>1.454274</td>\n",
       "      <td>11.914875</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.761038</td>\n",
       "      <td>0.121675</td>\n",
       "      <td>0.443863</td>\n",
       "      <td>0.333674</td>\n",
       "      <td>19.753078</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>1.494079</td>\n",
       "      <td>-0.205158</td>\n",
       "      <td>0.313068</td>\n",
       "      <td>-0.854096</td>\n",
       "      <td>23.607000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>95</th>\n",
       "      <td>-1.698106</td>\n",
       "      <td>0.387280</td>\n",
       "      <td>-2.255564</td>\n",
       "      <td>-1.022507</td>\n",
       "      <td>-21.564910</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>96</th>\n",
       "      <td>0.038631</td>\n",
       "      <td>-1.656715</td>\n",
       "      <td>-0.985511</td>\n",
       "      <td>-1.471835</td>\n",
       "      <td>-55.818011</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>97</th>\n",
       "      <td>1.648135</td>\n",
       "      <td>0.164228</td>\n",
       "      <td>0.567290</td>\n",
       "      <td>-0.222675</td>\n",
       "      <td>39.385820</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>98</th>\n",
       "      <td>-0.353432</td>\n",
       "      <td>-1.616474</td>\n",
       "      <td>-0.291837</td>\n",
       "      <td>-0.761492</td>\n",
       "      <td>-62.477272</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>99</th>\n",
       "      <td>0.857924</td>\n",
       "      <td>1.141102</td>\n",
       "      <td>1.466579</td>\n",
       "      <td>0.852552</td>\n",
       "      <td>56.572129</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>100 rows × 5 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "          x1        x2        x3        x4          y\n",
       "0   1.764052  0.400157  0.978738  2.240893  49.822907\n",
       "1   1.867558 -0.977278  0.950088 -0.151357   4.877286\n",
       "2  -0.103219  0.410599  0.144044  1.454274  11.914875\n",
       "3   0.761038  0.121675  0.443863  0.333674  19.753078\n",
       "4   1.494079 -0.205158  0.313068 -0.854096  23.607000\n",
       "..       ...       ...       ...       ...        ...\n",
       "95 -1.698106  0.387280 -2.255564 -1.022507 -21.564910\n",
       "96  0.038631 -1.656715 -0.985511 -1.471835 -55.818011\n",
       "97  1.648135  0.164228  0.567290 -0.222675  39.385820\n",
       "98 -0.353432 -1.616474 -0.291837 -0.761492 -62.477272\n",
       "99  0.857924  1.141102  1.466579  0.852552  56.572129\n",
       "\n",
       "[100 rows x 5 columns]"
      ]
     },
     "execution_count": 188,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_data"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "06592df0-5a46-4a07-a4fb-b87c789e8edd",
   "metadata": {},
   "source": [
    "**DataSet**\n",
    "\n",
    "Data sử dụng trong pytorch cần được chuyển về dạng Objects. Tối thiểu gồm 2 function căn bản :\n",
    "- `__len__` : Độ dài của dataset\n",
    "- `__getitem__` : Return về 1 sample data đã được convert qua dạng tensor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 189,
   "id": "7f666392-d3e4-4a5b-b01a-b71129314bd4",
   "metadata": {},
   "outputs": [],
   "source": [
    "class CustomDataset:\n",
    "    def __init__(self, data, target):\n",
    "        self.data = data\n",
    "        self.target = target\n",
    "    \n",
    "    def __len__(self):\n",
    "        \"\"\"\n",
    "        Return length of data\n",
    "        \"\"\"\n",
    "        return len(self.data)\n",
    "    \n",
    "    def __getitem__(self, idx):\n",
    "        \"\"\"\n",
    "        Return sample at idx as dictionary and data type is tensor\n",
    "        \"\"\"\n",
    "        current_sample = self.data[idx, :]\n",
    "        current_target = self.target[idx]\n",
    "        return {\n",
    "            \"sample\": torch.tensor(current_sample, dtype=torch.float),\n",
    "            \"target\": torch.tensor(current_target, dtype=torch.float)\n",
    "        }\n",
    "    \n",
    "    def to_pandas(self, features, target):\n",
    "        \"\"\"\n",
    "        Not required, this function used to visualize predict values\n",
    "        \"\"\"\n",
    "        df = pd.DataFrame(self.data, columns=features)\n",
    "        df[target] = self.target\n",
    "        \n",
    "        return df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 127,
   "id": "2340a5d1-3f16-4645-aefb-4df651e76587",
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset = CustomDataset(data=X, target=y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 129,
   "id": "deacc30b-6d11-4035-8a05-843ef89b5f43",
   "metadata": {},
   "outputs": [],
   "source": [
    "dataloader = DataLoader(dataset, batch_size=8)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2cfd661d-67a7-446c-9c65-eb15e0120a95",
   "metadata": {},
   "source": [
    "### 4.2 Build model\n",
    "\n",
    "- Loss Function sử dụng là MSE. https://pytorch.org/docs/stable/generated/torch.nn.MSELoss.html#torch.nn.MSELoss"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 191,
   "id": "ebe7fd74-20b6-45c3-8f5f-fc5d9b603b87",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Define model \n",
    "\n",
    "model = lambda x, w, b: torch.matmul(x, w) + b"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 193,
   "id": "290b75d9-6ba9-4f97-8e93-4a99ac20c025",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch :  0  Loss :  20441.014770507812\n",
      "Epoch :  10  Loss :  11744.809295654297\n",
      "Epoch :  20  Loss :  6913.106582641602\n",
      "Epoch :  30  Loss :  4175.6396484375\n",
      "Epoch :  40  Loss :  2589.5048599243164\n",
      "Epoch :  50  Loss :  1647.4812316894531\n",
      "Epoch :  60  Loss :  1073.266757965088\n",
      "Epoch :  70  Loss :  714.0120697021484\n",
      "Epoch :  80  Loss :  483.58666801452637\n",
      "Epoch :  90  Loss :  332.403281211853\n"
     ]
    }
   ],
   "source": [
    "# Init weight and b\n",
    "w = torch.randn(4, 1, requires_grad=True)\n",
    "b = torch.randn(1, requires_grad=True)\n",
    "\n",
    "learning_rate = 0.001\n",
    "eporchs = 100\n",
    "\n",
    "for epoch in range(eporchs):\n",
    "        \n",
    "    epoch_loss = 0\n",
    "    for batch in dataloader:\n",
    "        _x = batch[\"sample\"]\n",
    "        _y = batch[\"target\"].reshape(-1,1)\n",
    "        \n",
    "        output = model(_x, w, b)\n",
    "        \n",
    "        loss = torch.nn.MSELoss()(output, _y)\n",
    "        \n",
    "        loss.backward()\n",
    "        \n",
    "        # Optimizer\n",
    "        with torch.no_grad():\n",
    "            w = w - learning_rate * w.grad\n",
    "            b = b - learning_rate * b.grad\n",
    "            \n",
    "        w.requires_grad_(True)\n",
    "        b.requires_grad_(True)\n",
    "        \n",
    "        epoch_loss += loss.item()\n",
    "        \n",
    "    if epoch % 10 == 0:\n",
    "        print(\"Epoch : \",epoch, \" Loss : \",epoch_loss)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 194,
   "id": "f543519d-a3c1-4d43-8b0f-533eef1be7ac",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Weight of w after learning  tensor([[18.5350],\n",
      "        [30.4026],\n",
      "        [ 2.9158],\n",
      "        [ 0.9457]], requires_grad=True) . Bias =  tensor([-0.2636], requires_grad=True)\n"
     ]
    }
   ],
   "source": [
    "print(\"Weight of w after learning \", w, \". Bias = \", b)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7caf7b19-6c1f-4d2d-8226-87cc02ec7a3a",
   "metadata": {},
   "source": [
    "### 4.3 Visualize predict"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e8b22e14-303d-41cf-b123-5a858d144b16",
   "metadata": {},
   "source": [
    "**X : Must be a tensor. We used customDataset to make each sample become a tensor**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 195,
   "id": "6bccc8b2-8b5b-4a16-afaf-b4af56503e20",
   "metadata": {},
   "outputs": [],
   "source": [
    "predict_dataset = CustomDataset(df_data[[\"x1\", \"x2\", \"x3\", \"x4\"]].values, df_data[\"y\"].values)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 196,
   "id": "8a767011-abad-48c1-bacf-798512dca6f2",
   "metadata": {},
   "outputs": [],
   "source": [
    "y_predict = []\n",
    "\n",
    "for sample in predict_dataset:\n",
    "    _x = sample[\"sample\"]\n",
    "    _y_predict = model(_x, w, b)\n",
    "    \n",
    "    y_predict.append(_y_predict.tolist()[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 197,
   "id": "3353b863-6164-490e-8699-c908a8822562",
   "metadata": {},
   "outputs": [],
   "source": [
    "df_predict = predict_dataset.to_pandas(features=[\"x1\",  \"x2\", \"x3\", \"x4\"], target=\"y\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 198,
   "id": "88b2c0f5-62f0-45bc-9f9c-ba65e5fcbd7a",
   "metadata": {},
   "outputs": [],
   "source": [
    "df_predict[\"predict\"] = y_predict"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 199,
   "id": "04c43741-0434-40e5-8f44-59c45d87248d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<AxesSubplot:>"
      ]
     },
     "execution_count": 199,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "df_predict[[\"y\", \"predict\"]].plot()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "643698df-55ba-4d70-9ad5-db97b97f2901",
   "metadata": {},
   "source": [
    "# II. Hardware"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "6e8b4196-cf70-48a0-b26a-f4a4ef4dba80",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7e072a8b-d78a-4421-bbc9-1112e20c5291",
   "metadata": {},
   "source": [
    "## 1. Driver check"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bbebd49a-14a8-48a1-bded-bc6aff4e8927",
   "metadata": {},
   "source": [
    "**Kiểm tra GPU Nvdia driver đã được cài đặt chưa**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "f8d67fd4-959e-4c56-afc4-288485082446",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "True\n"
     ]
    }
   ],
   "source": [
    "print(torch.cuda.is_available())"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "48bff42a-ee2c-4552-b846-ad31c179bb30",
   "metadata": {},
   "source": [
    "**Toàn bộ tensor đều được khởi tạo mặc định trên CPU.**\n",
    "\n",
    "> [variable].is_cuda `# Check 1 variable có được mặc định tính toán trên GPU không`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "eb80316e-730e-4e13-8e9e-6f77b242a79f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "False\n"
     ]
    }
   ],
   "source": [
    "x1 = torch.tensor([0.1, 0.2, 0.3])\n",
    "print(x1.is_cuda)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ee74fb68-b36a-48cc-87cd-bc381259e98f",
   "metadata": {},
   "source": [
    "Cách check model chạy trên CPU or GPU."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "ff5e6978-7085-4f89-87b4-03f5eac84c6e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "False\n"
     ]
    }
   ],
   "source": [
    "model = torch.nn.Linear(5, 1)\n",
    "print(next(model.parameters()).is_cuda)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "248de5d0-de2c-4740-bae3-8428e4be4e62",
   "metadata": {},
   "source": [
    "## 2. Convert CPU to GPU"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "251eacb4-2f92-4c5d-8b9c-580495590b9e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "cuda:0\n"
     ]
    }
   ],
   "source": [
    "# Define divice name. Nếu GPU có thể sử dụng, device sẽ có tên cuda:0\n",
    "device = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n",
    "print(device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "f6071796-7ce9-43e7-b551-3ab5cc6bd7e1",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Is variable run on GPU :  True\n",
      "Is model run on GPU : True\n"
     ]
    }
   ],
   "source": [
    "x1 = x1.to(device)\n",
    "print(\"Is variable run on GPU : \", x1.is_cuda)\n",
    "\n",
    "model = model.to(device)\n",
    "print(\"Is model run on GPU :\", next(model.parameters()).is_cuda)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "18e2d46e-df75-4c81-98c7-a70191624d59",
   "metadata": {},
   "source": [
    "*Tensor cuda package có hỗ trợ function convert qua gpu*"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "id": "5b3e065e-1f82-470a-b962-c8ded67a5bfc",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Is x2 run on GPU :  True\n"
     ]
    }
   ],
   "source": [
    "x2 = torch.tensor([0.1, 0.2])\n",
    "x2 = x2.cuda()\n",
    "print(\"Is x2 run on GPU : \", x2.is_cuda)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "id": "00a7ccf3-dc7b-48fd-84c8-330042d6d014",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of device :  1\n",
      "Current using index of device :  0\n",
      "Device name is :  NVIDIA GeForce RTX 3070 Laptop GPU\n",
      "Device memory is :  2097152\n",
      "Memory using is :  3072\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\Onsiter\\.conda\\envs\\nlp\\lib\\site-packages\\torch\\cuda\\memory.py:386: FutureWarning: torch.cuda.memory_cached has been renamed to torch.cuda.memory_reserved\n",
      "  FutureWarning)\n"
     ]
    }
   ],
   "source": [
    "print(\"Number of device : \", torch.cuda.device_count())\n",
    "\n",
    "print(\"Current using index of device : \", torch.cuda.current_device())\n",
    "\n",
    "print(\"Device name is : \", torch.cuda.get_device_name(0))\n",
    "\n",
    "print(\"Device memory is : \", torch.cuda.memory_cached())\n",
    "\n",
    "print(\"Memory using is : \", torch.cuda.memory_allocated())"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c174094d-c5df-41b1-ada9-22458d26f5ab",
   "metadata": {},
   "source": [
    "Clear GPU cache"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "id": "a13116d5-577c-4ddf-ac29-6af9e7f2bb7a",
   "metadata": {},
   "outputs": [],
   "source": [
    "torch.cuda.empty_cache()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "96010b09-e6a0-4d06-9ae0-047cfb2f89f3",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.7.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}