Python29-EffectiveTensorflow.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Tensorflow基础"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "Tensorflow与其他数值计算库最大的区别就是操作符号化。这个强大的概念使得Tensorflow可以处理各种事情（例如自动区分），而其他库无法实现这些功能（例如numpy）\n",
    "\n",
    "我们以两个矩阵相乘为例，首先我们看看numpy如何实现"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[ -2.92561259  -7.55579319   2.76137022  -1.33024061  -2.19115835\n",
      "    1.60398038  -4.93829921   3.69232293   2.64791158   3.86324054]\n",
      " [ -3.00912398  -2.18463465   1.89091331   1.45675056   1.13008171\n",
      "    0.78199161  -2.46933188   3.54975584   0.92857783   2.87275187]\n",
      " [  4.87175962   0.26460832  -3.78383175   2.80412933  -0.25617869\n",
      "    4.01327401   3.05237345  -8.99186896   1.79235241  -4.86715496]\n",
      " [ -1.74415036   1.58789346  -0.90739707   0.88665551   1.38826095\n",
      "    2.70925499  -0.51533716  -2.17419073  -0.72661957   0.29646348]\n",
      " [  2.18615508  -1.56540798   1.39808239  -2.78030005   0.75153573\n",
      "    0.05218568  -0.41075664   0.08834171  -2.87388832   2.23081798]\n",
      " [ -3.03344325  -3.85994409  -1.48369781   3.96432509   2.10302034\n",
      "    4.22844567  -1.60628181   0.81367515  -1.12597991   3.30366936]\n",
      " [  1.00651427 -10.18377206  -2.52774076   0.51471813  -3.58806337\n",
      "    0.77538835  -4.85166005   1.53741656   7.04761688   1.792273  ]\n",
      " [  2.01466502   4.16162317   7.07641281  -1.21377464  -4.72875263\n",
      "    2.7748531    4.44133838  -2.53660738  -6.8104239   -1.01322609]\n",
      " [ -0.31376823   3.33142982  -0.57210342   0.14533245   2.12745336\n",
      "   -1.15558262   0.45203087   0.21946917   0.53338409  -3.0755217 ]\n",
      " [  1.8085264   -1.6867176   -0.665157     1.79754705  -0.04868972\n",
      "   -3.01805074   4.57167437   1.70637805   2.64346387   0.75928498]]\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "\n",
    "x=np.random.normal(size=[10,10])\n",
    "y=np.random.normal(size=[10,10])\n",
    "z=np.dot(x,y)\n",
    "print(z)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "接着我们使用Tensorflow实现同样的计算"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensor(\"MatMul_4:0\", shape=(10, 10), dtype=float32)\n",
      "[[ -3.26649714e+00  -7.61130691e-01   6.06034338e-01  -1.26068842e+00\n",
      "    1.74175835e+00   6.50460958e-01   2.48260915e-01  -2.05012381e-01\n",
      "   -3.29370213e+00  -1.03132117e+00]\n",
      " [  1.81942987e+00  -2.06936169e+00   2.06779933e+00  -3.66959906e+00\n",
      "    4.36944532e+00   3.03501344e+00   4.72806644e+00  -3.20354390e+00\n",
      "   -3.32232356e+00  -4.45390654e+00]\n",
      " [ -4.22742546e-01   2.37801933e+00  -4.84825945e+00  -1.64431429e+00\n",
      "    2.14773345e+00  -5.46641648e-02   2.37222910e-02  -1.68210483e+00\n",
      "   -5.92576087e-01  -2.30085659e+00]\n",
      " [  4.57029724e+00  -1.32026494e+00  -2.27896333e+00  -2.30306816e+00\n",
      "    7.42745399e-03  -3.94744158e+00  -1.28421330e+00   2.22644424e+00\n",
      "    4.51918650e+00   1.45548272e+00]\n",
      " [  7.57214499e+00  -6.07138586e+00  -1.05868793e+00   5.87156892e-01\n",
      "    1.21003914e+00  -2.44661140e+00  -2.75172830e+00  -7.21841455e-01\n",
      "    6.16718578e+00  -3.27266359e+00]\n",
      " [  2.57871008e+00  -2.04265738e+00   5.80107689e-01  -1.07904172e+00\n",
      "   -2.82521152e+00   1.83521628e+00  -8.31910849e-01   3.99815130e+00\n",
      "    1.81486487e-01   2.59252381e+00]\n",
      " [  1.11718488e+00  -1.12849116e-01   2.66168141e+00  -3.66602015e+00\n",
      "   -5.52139091e+00   5.17316151e+00   4.41561604e+00  -4.40914631e+00\n",
      "   -7.55649853e+00  -1.08792572e+01]\n",
      " [  1.94527221e+00  -2.03468347e+00   1.87479949e+00  -3.73102617e+00\n",
      "   -2.78824186e+00   2.49608350e+00   1.45620012e+00   6.01150894e+00\n",
      "   -1.75918555e+00   2.33199430e+00]\n",
      " [  3.43171287e+00   3.61559439e+00  -1.52311206e+00   4.07898664e+00\n",
      "   -8.72540593e-01  -5.22466063e-01   2.32854509e+00  -1.53407001e+00\n",
      "    3.62039042e+00   2.32134366e+00]\n",
      " [ -1.44251347e+00  -2.02366018e+00  -6.42121494e-01  -2.56414175e-01\n",
      "    2.48252082e+00   4.88413048e+00   1.23599446e+00   4.04822397e+00\n",
      "   -7.04676747e-01   4.08192694e-01]]\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "\n",
    "x=tf.random_normal([10,10])\n",
    "y=tf.random_normal([10,10])\n",
    "z=tf.matmul(x,y)\n",
    "sess=tf.Session()\n",
    "\n",
    "z_val=sess.run(z)\n",
    "\n",
    "print(z_val)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "不像numpy那样立即进行相乘并产生结果，Tensorflow仅给我们一个节点（在图中）的句柄，如果我们直接打印“z”,会出现下面的结果"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensor(\"MatMul_4:0\", shape=(10, 10), dtype=float32)\n"
     ]
    }
   ],
   "source": [
    "print(z)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "因为两个输入(x和y)有了明确定义的形状（shape）,tensorflow能够推断出张量的形状以及它的类型，为了计算出张量的值，我们需要创建一个会话（session），然后使用Session.run()去评估"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "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": "3.6.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}