5_sgd_test_theano_進階版.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# sgd theano 進階版寫法\n",
    "\n",
    "首先假設一個很簡單的function => y=x\n",
    "\n",
    "1. 模擬資料一些資料從y=x產生\n",
    "2. init weights\n",
    "3. define model，真實世界裡的 <span style='color:red'>f</span> 我們是未知的，這裡我們當作已經知道是y=x，方便用來驗證gradient\n",
    "4. define cost\n",
    "5. implement SGD\n",
    "6. training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Populating the interactive namespace from numpy and matplotlib\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Anaconda2\\lib\\site-packages\\IPython\\core\\magics\\pylab.py:161: UserWarning: pylab import has clobbered these variables: ['f']\n",
      "`%matplotlib` prevents importing * from pylab and numpy\n",
      "  \"\\n`%matplotlib` prevents importing * from pylab and numpy\"\n"
     ]
    }
   ],
   "source": [
    "%pylab inline\n",
    "import theano\n",
    "import theano.tensor as T\n",
    "#import numpy as np\n",
    "#import matplotlib.pyplot as plt\n",
    "import itertools as it"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 把floatX包起來方便使用"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def floatX(X):\n",
    "    return np.asarray(X, dtype=theano.config.floatX)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 模擬資料"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "x_data = floatX(np.array(np.linspace(-5.0, 5.0, 100)))\n",
    "y_data = floatX(np.array(x_data))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 初始化權重"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "w = theano.shared(-1.)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 定義model / function\n",
    "\n",
    "這裡要特別注意，因為是跑SGD是一筆一筆資料跑而且我們資料只有一維，要將x宣告為theano <span style='color:red'>scalar()</span>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "### model\n",
    "x = T.scalar()\n",
    "y = w*x"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "如果要測試model output可以theano function一下"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "-10.0\n"
     ]
    }
   ],
   "source": [
    "f = theano.function([x], y)\n",
    "print f(10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "那如果x是二維或是多維以上呢?那就要將x宣告為theano <span style='color:red'>vector()</span>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "w2 我是vector\n",
      "<TensorType(float64, vector)>\n",
      "\n",
      "x2 我也是vector\n",
      "<TensorType(float64, vector)>\n",
      "\n",
      "Element Wise 相乘，我還是vector\n",
      "[-1. -1.]\n",
      "<TensorType(float64, vector)>\n",
      "\n",
      "矩陣相乘dot，我變成scalar了，這樣才能配合y_hat也是scalar，順利做gradient\n",
      "-2.0\n",
      "<TensorType(float64, scalar)>\n"
     ]
    }
   ],
   "source": [
    "#假設x是二維，w也要是二維(當然)\n",
    "w2 = theano.shared(floatX(np.array([-1.,-1.])))\n",
    "print \"w2 我是vector\"\n",
    "print w2\n",
    "print \"\"\n",
    "x2 = theano.shared(floatX(np.array([1.,1.])))\n",
    "print \"x2 我也是vector\"\n",
    "print x2\n",
    "print \"\"\n",
    "print \"Element Wise 相乘，我還是vector\"\n",
    "print (w2*x2).eval()\n",
    "print (w2*x2).type()\n",
    "print \"\"\n",
    "print \"矩陣相乘dot，我變成scalar了，這樣才能配合y_hat也是scalar，順利做gradient\"\n",
    "print (T.dot(w2,x2)).eval()\n",
    "print (T.dot(w2,x2)).type()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 實作SGD\n",
    "* 用izip配合迴圈寫法一次update多個參數，如同下面寫法\n",
    "* [w1, w1 - 0.01*dw1]\n",
    "* [w2, w2 - 0.01*dw2]\n",
    "* ...\n",
    "* [b, b - 0.01*db]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "### gradients\n",
    "def sgd(params, grads, lr): \n",
    "    updates = []\n",
    "    for p, g in it.izip(params, grads):      \n",
    "        updates.append([p, p - lr*g])    \n",
    "    return updates"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 定義cost/error/loss\n",
    "#### 很重要，取決你怎麼餵資料算cost，SGD/MiniBatch/All Data cost算法都不一樣)\n",
    "\n",
    "要特別注意cost function是<span style='color:red'>不用做sum</span>的(別照抄把全部data看過一輪的cost function)，因為是一筆做一次gradient"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "### cost/error/loss\n",
    "y_hat = T.scalar()\n",
    "#cost = T.sum((y-y_hat)**2)\n",
    "cost = (y-y_hat)**2"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 定義train的theano function\n",
    "* 對cost的w偏微分\n",
    "* 輸入x, y_hat\n",
    "* 輸出cost, w\n",
    "* 更新用sgd"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "dw = T.grad(cost, w)\n",
    "train = theano.function(inputs=[x, y_hat],\n",
    "                        outputs=[cost,w],\n",
    "                        updates=sgd([w], [dw], 0.001))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 洗牌一下"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# shuffle idx for Stochastic (不用作也沒關係，有可能data本身就是沒有順序性了)\n",
    "idx = np.arange(x_data.shape[0])\n",
    "np.random.shuffle(idx)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 開始training\n",
    "\n",
    "* cost_result第一個是cost變化，第二個是w的變化"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[array(0.23693878960409676), array(0.6430400362188393)]\n",
      "[array(0.007491674171752809), array(0.9365267309182803)]\n",
      "[array(0.00023687629192961854), array(0.9887134236421247)]\n",
      "[array(7.489698082423297e-06), array(0.9979930637302116)]\n"
     ]
    }
   ],
   "source": [
    "# training\n",
    "cost_list = []\n",
    "for t in range(4):    \n",
    "    for i in idx:\n",
    "        cost_result = train(x_data[i], y_data[i])        \n",
    "        cost_list.append(cost_result[0])\n",
    "    print cost_result"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 畫圖看結果\n",
    "\n",
    "* SGD不是像GD一樣是看最陡的方向，而是取一筆後就走了，會亂跳，但是會慢慢收斂\n",
    "* SGD比GD快(但是如果在<span style='color:red'>GPU平行運算</span>的環境下，反而是mini batch或all data比較快，因為矩陣相乘靠GPU平行運算下 (ref ML2016 Hello World of deep learning)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAh4AAAF5CAYAAADQ2iM1AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nRK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0xc3be8d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(cost_list)\n",
    "plt.xlabel(\"No. of parameters updates\")\n",
    "plt.ylabel(\"Loss/Cost/Error\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 如果不做shuffle呢???"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[array(3.5034466188051168), array(0.6256500771307617)]\n",
      "[array(0.1107740973526242), array(0.9334345142427025)]\n",
      "[array(0.0035025225098115694), array(0.9881635773819764)]\n",
      "[array(0.00011074487831468541), array(0.9978952921503003)]\n"
     ]
    }
   ],
   "source": [
    "#把w重新初始化\n",
    "w.set_value(-1.)\n",
    "\n",
    "idx = np.arange(x_data.shape[0])\n",
    "\n",
    "# training\n",
    "cost_list = []\n",
    "for t in range(4):    \n",
    "    for i in idx:\n",
    "        cost_result = train(x_data[i], y_data[i])        \n",
    "        cost_list.append(cost_result[0])\n",
    "    print cost_result"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 結果好平滑，為什麼?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAicAAAF5CAYAAABEPIrHAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0xc4d5ba8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(cost_list)\n",
    "plt.xlabel(\"No. of parameters updates\")\n",
    "plt.ylabel(\"Loss/Cost/Error\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 因為我們cost function是長這樣(convex)，我們是從最左邊慢慢往下走，而不是shuffle亂走，結果就會很平滑\n",
    "- 把w重新初始化-1，實際f的w=1，這樣就能畫出每一筆x計算出square error的結果"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAicAAAFkCAYAAAD7dJuCAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nVJyIiIhIpPwfrlI4Vc8uLDUAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0xc462438>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "w.set_value(-1.)\n",
    "\n",
    "cost_f_list = []\n",
    "cost_f = theano.function([x, y_hat], cost)\n",
    "for i in range(x_data.shape[0]):\n",
    "    cost_f_list.append(cost_f(x_data[i], y_data[i]))\n",
    "    \n",
    "plt.plot(cost_f_list)\n",
    "plt.ylabel(\"Loss/Cost/Error\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python [conda root]",
   "language": "python",
   "name": "conda-root-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}