5.1卷积神经网络之二维卷积层.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.0.0\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "import os\n",
    "\n",
    "print(tf.__version__)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 自己实现一个卷积运算函数,输入时初始数组和卷积核数组,输出卷积计算后的结果\n",
    "\n",
    "def conv2d(X, K):\n",
    "    h, w = K.shape\n",
    "    # 先计算出输出结果 y 的形状，构造y，方便计算\n",
    "    Y = tf.Variable(tf.zeros((X.shape[0] - h + 1, X.shape[1] - w +1)))\n",
    "    for i in range(Y.shape[0]):\n",
    "        for j in range(Y.shape[1]):\n",
    "            # 将 y 的 i,j 赋值成 和\n",
    "            Y[i,j].assign(tf.cast(tf.reduce_sum(X[i:i+h, j:j+w] * K), dtype=tf.float32))\n",
    "    return Y\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Variable 'Variable:0' shape=(2, 2) dtype=float32, numpy=\n",
       "array([[19., 25.],\n",
       "       [37., 43.]], dtype=float32)>"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X = tf.constant([[0,1,2], [3,4,5], [6,7,8]])\n",
    "K = tf.constant([[0,1], [2,3]])\n",
    "conv2d(X, K)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 构造卷积层， conv2d + bias\n",
    "\n",
    "class Conv2D(tf.keras.layers.Layer):\n",
    "    def __init__(self):\n",
    "        super().__init__()\n",
    "    \n",
    "    # 由于有 训练的参数，所以使用 build构建\n",
    "    def build(self, kernal_size):\n",
    "        self.w = self.add_weight(name='weight',shape=kernal_size,initializer=tf.random_normal_initializer())\n",
    "        self.b = self.add_weight(name='bias',shape=kernal_size,initializer=tf.zeros_initializer())\n",
    "    def call(self, inputs):\n",
    "        y = conv2d(inputs, self.w) + self.b\n",
    "        \n",
    "        return y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Variable 'UnreadVariable' shape=(6, 8) dtype=float32, numpy=\n",
       "array([[1., 1., 0., 0., 0., 0., 1., 1.],\n",
       "       [1., 1., 0., 0., 0., 0., 1., 1.],\n",
       "       [1., 1., 0., 0., 0., 0., 1., 1.],\n",
       "       [1., 1., 0., 0., 0., 0., 1., 1.],\n",
       "       [1., 1., 0., 0., 0., 0., 1., 1.],\n",
       "       [1., 1., 0., 0., 0., 0., 1., 1.]], dtype=float32)>"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 构建数据\n",
    "X = tf.Variable(tf.ones((6,8)))\n",
    "X[:, 2:6].assign(tf.zeros(X[:,2:6].shape))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Variable 'Variable:0' shape=(6, 7) dtype=float32, numpy=\n",
       "array([[ 0.,  1.,  0.,  0.,  0., -1.,  0.],\n",
       "       [ 0.,  1.,  0.,  0.,  0., -1.,  0.],\n",
       "       [ 0.,  1.,  0.,  0.,  0., -1.,  0.],\n",
       "       [ 0.,  1.,  0.,  0.,  0., -1.,  0.],\n",
       "       [ 0.,  1.,  0.,  0.,  0., -1.,  0.],\n",
       "       [ 0.,  1.,  0.,  0.,  0., -1.,  0.]], dtype=float32)>"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "K = tf.constant([[1,-1]], dtype = tf.float32)\n",
    "Y = conv2d(X, K)\n",
    "Y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "X = tf.reshape(X, (1, 6, 8, 1))\n",
    "Y = tf.reshape(Y, (1,6,7,1))\n",
    "\n",
    "conv2d = tf.keras.layers.Conv2D(1, (1,2))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "batch 2, loss 0.210\n",
      "batch 4, loss 0.086\n",
      "batch 6, loss 0.035\n",
      "batch 8, loss 0.014\n",
      "batch 10, loss 0.006\n",
      "batch 12, loss 0.002\n",
      "batch 14, loss 0.001\n",
      "batch 16, loss 0.000\n",
      "batch 18, loss 0.000\n",
      "batch 20, loss 0.000\n"
     ]
    }
   ],
   "source": [
    "Y_hat = conv2d(X)\n",
    "for i in range(20):\n",
    "    with tf.GradientTape(watch_accessed_variables=False) as g:\n",
    "        g.watch(conv2d.weights[0])\n",
    "        Y_hat = conv2d(X)\n",
    "        l = (abs(Y_hat - Y)) ** 2\n",
    "        dl = g.gradient(l, conv2d.weights[0])\n",
    "        lr = 3e-2\n",
    "        update = tf.multiply(lr, dl)\n",
    "        updated_weights = conv2d.get_weights()\n",
    "        updated_weights[0] = conv2d.weights[0] - update\n",
    "        conv2d.set_weights(updated_weights)  \n",
    "\n",
    "        if (i + 1)% 2 == 0:\n",
    "            print('batch %d, loss %.3f' % (i + 1, tf.reduce_sum(l)))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<tf.Tensor: id=3352, shape=(1, 2), dtype=float32, numpy=array([[ 0.9991463, -1.0008522]], dtype=float32)>"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tf.reshape(conv2d.get_weights()[0],(1,2))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python (tf2)",
   "language": "python",
   "name": "tf2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}