Day95-CNN_Pooling_Padding_HW.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 作業目標:\n",
    "    了解 Padding 的作法\n",
    "    \n",
    "    了解 Pooling 的作法"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# 作業重點:\n",
    "    (1) 調整Padding 的設定, 看 Pad 後的結果\n",
    "    \n",
    "    (2) 調整Pooling 的設定, 看 Poolin 後的結果"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The autoreload extension is already loaded. To reload it, use:\n",
      "  %reload_ext autoreload\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "import h5py\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "# 繪圖結果直接顯示在Jupyter cell 之內\n",
    "%matplotlib inline  \n",
    "plt.rcParams['figure.figsize'] = (5.0, 4.0) #  設定繪圖板的大小\n",
    "plt.rcParams['image.interpolation'] = 'nearest'\n",
    "plt.rcParams['image.cmap'] = 'gray'\n",
    "\n",
    "# autoreload。可以讓我們不退出IPython就動態修改代碼，在執行代碼前IPython會幫我們自動重載改動的模塊\n",
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "np.random.seed(1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# GRADED FUNCTION: zero_pad\n",
    "def zero_pad(X, pad):\n",
    "    \"\"\"\n",
    "    對image X 做 zero-padding. \n",
    "    參數定義如下:\n",
    "    X -- python numpy array, 呈現維度 (m, n_H, n_W, n_C), 代表一批 m 個圖像\n",
    "         n_H: 圖高, n_W: 圖寬, n_C: color channels 數\n",
    "    pad -- 整數, 加幾圈的 zero padding.\n",
    "    Returns:\n",
    "    X_pad -- image of shape (m, n_H + 2*pad, n_W + 2*pad, n_C) 做完zero-padding 的結果\n",
    "    \"\"\"\n",
    "    \n",
    "    X_pad = np.pad(X, ((0, 0), (pad, pad), (pad, pad), (0, 0)), 'constant', constant_values=(0, 0))\n",
    "  \n",
    "    return X_pad"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "x.shape = (4, 3, 3, 2)\n",
      "x_pad.shape = (4, 9, 9, 2)\n",
      "x[1,1] = [[ 0.90085595 -0.68372786]\n",
      " [-0.12289023 -0.93576943]\n",
      " [-0.26788808  0.53035547]]\n",
      "x_pad[1,1] = [[0. 0.]\n",
      " [0. 0.]\n",
      " [0. 0.]\n",
      " [0. 0.]\n",
      " [0. 0.]\n",
      " [0. 0.]\n",
      " [0. 0.]\n",
      " [0. 0.]\n",
      " [0. 0.]]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x1d26bca3a58>"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 360x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "np.random.seed(1)\n",
    "\n",
    "x = np.random.randn(4, 3, 3, 2) #產生gray image\n",
    "x_pad = zero_pad(x, 3) # 加兩圈 Pad\n",
    "\n",
    "print (\"x.shape =\", x.shape)\n",
    "print (\"x_pad.shape =\", x_pad.shape)\n",
    "print (\"x[1,1] =\", x[1,1])\n",
    "print (\"x_pad[1,1] =\", x_pad[1,1])\n",
    "fig, axarr = plt.subplots(1, 2)\n",
    "axarr[0].set_title('x')\n",
    "axarr[0].imshow(x[0,:,:,0])\n",
    "axarr[1].set_title('x_pad')\n",
    "axarr[1].imshow(x_pad[0,:,:,0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "# GRADED FUNCTION: pool_forward\n",
    "def pool_forward(A_prev, hparameters, mode = \"max\"):\n",
    "    \"\"\"\n",
    "    設計一個前行網路的池化層\n",
    "    參數定義如下:\n",
    "    A_prev -- 輸入的numpy 陣列, 維度 (m, n_H_prev, n_W_prev, n_C_prev)\n",
    "    hparameter 超參數 --  \"f\" and \"stride\" 所形成的python 字典\n",
    "    mode -- 池化的模式: \"max\" or \"average\"\n",
    "    \n",
    "    返回:\n",
    "        A -- 輸出的池化層, 維度為 (m, n_H, n_W, n_C) 的 numpy 陣列\n",
    "        cache -- 可以應用在 backward pass pooling layer 資料, 包含 input and hparameter\n",
    "    \"\"\"\n",
    "\n",
    "    # 檢索尺寸 from the input shape\n",
    "    (m, n_H_prev, n_W_prev, n_C_prev) = A_prev.shape\n",
    " \n",
    "    # 檢索超參數 from \"hparameters\"\n",
    "    f = hparameters[\"f\"]\n",
    "    stride = hparameters[\"stride\"]\n",
    "\n",
    "    # 定義輸出的dimensions\n",
    "    n_H = int(1 + (n_H_prev - f) / stride)\n",
    "    n_W = int(1 + (n_W_prev - f) / stride)\n",
    "    n_C = n_C_prev\n",
    "\n",
    "    # 初始化輸出的 matrix A\n",
    "    A = np.zeros((m, n_H, n_W, n_C))\n",
    "\n",
    "    ### 程式起始位置  ###\n",
    "    for i in range(m): # 訓練樣本的for 迴圈\n",
    "        for h in range(n_H): # 輸出樣本的for 迴圈, 針對vertical axis\n",
    "            for w in range(n_W): #  輸出樣本的for 迴圈, 針對 horizontal axis\n",
    "                for c in range (n_C): #  輸出樣本的for 迴圈, 針對channels\n",
    "\n",
    "                     # 找出特徵圖的寬度跟高度四個點\n",
    "                    vert_start = h * stride\n",
    "                    vert_end = h * stride+ f\n",
    "                    horiz_start = w * stride\n",
    "                    horiz_end = w * stride + f\n",
    "\n",
    "                    # 定義第i個訓練示例\n",
    "                    a_prev_slice = A_prev[i, vert_start:vert_end, horiz_start:horiz_end,c]\n",
    "\n",
    "                    # 計算輸入data 的池化結果. 使用 if statment 去做分類\n",
    "                    if mode == \"max\":\n",
    "                        A[i, h, w, c] = np.max(a_prev_slice)\n",
    "                    elif mode == \"average\":\n",
    "                        A[i, h, w, c] = np.mean(a_prev_slice)\n",
    "\n",
    "                        ### 程式結束  ###\n",
    "    \n",
    "    # 儲存輸入的特徵圖跟所設定的超參數, 可以用在 pool_backward()\n",
    "    cache = (A_prev, hparameters)\n",
    "    \n",
    "    # 確認輸出的資料維度\n",
    "    assert(A.shape == (m, n_H, n_W, n_C))\n",
    "    return A, cache"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "mode = max\n",
      "A = [[[[ 1.62434536 -0.61175641 -0.52817175]\n",
      "   [-0.24937038  1.46210794 -2.06014071]]\n",
      "\n",
      "  [[-1.11731035  0.2344157   1.65980218]\n",
      "   [-0.63699565  0.19091548  2.10025514]]]\n",
      "\n",
      "\n",
      " [[[ 0.12015895  0.61720311  0.30017032]\n",
      "   [ 0.93110208  0.28558733  0.88514116]]\n",
      "\n",
      "  [[ 0.19829972  0.11900865 -0.67066229]\n",
      "   [-0.63873041  0.42349435  0.07734007]]]]\n",
      "\n",
      "mode = average\n",
      "A = [[[[ 1.62434536 -0.61175641 -0.52817175]\n",
      "   [-0.24937038  1.46210794 -2.06014071]]\n",
      "\n",
      "  [[-1.11731035  0.2344157   1.65980218]\n",
      "   [-0.63699565  0.19091548  2.10025514]]]\n",
      "\n",
      "\n",
      " [[[ 0.12015895  0.61720311  0.30017032]\n",
      "   [ 0.93110208  0.28558733  0.88514116]]\n",
      "\n",
      "  [[ 0.19829972  0.11900865 -0.67066229]\n",
      "   [-0.63873041  0.42349435  0.07734007]]]]\n"
     ]
    }
   ],
   "source": [
    "np.random.seed(1)\n",
    "A_prev = np.random.randn(2, 4, 4, 3)\n",
    "hparameters = {\"stride\" : 3, \"f\": 1}\n",
    "\n",
    "A, cache = pool_forward(A_prev, hparameters)\n",
    "print(\"mode = max\")\n",
    "print(\"A =\", A)\n",
    "print()\n",
    "A, cache = pool_forward(A_prev, hparameters, mode = \"average\")\n",
    "print(\"mode = average\")\n",
    "print(\"A =\", A)"
   ]
  },
  {
   "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": "3.6.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}