AlexNet.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "from network import *\n",
    "\n",
    "import pickle"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Local Response Normalization\n",
    "\\begin{equation*}\n",
    "{ b }_{ x,y }^{ i }={ a }_{ x,y }^{ i }/{ \\left( k+\\alpha \\sum _{ j=max(0,i-n/2) }^{ min(N-1,i+n/2) }{ { ({ a }_{ x,y }^{ j }) }^{ 2 } }  \\right)  }^{ \\beta  }\n",
    "\\end{equation*}\n",
    "k = 2, n = 5, alpha = 10^-4, beta = 0.75\n",
    "\n",
    "conv 1, 2의 ReLU한 결과에 적용"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "class LRN:\n",
    "    def __init__(self):\n",
    "        pass\n",
    "    \n",
    "    def forward(self, x):\n",
    "        pass\n",
    "    \n",
    "    def backward(self, dout):\n",
    "        pass\n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Data Augmentation\n",
    "256x256 crop random 224x224 5개<br/>\n",
    "좌우 반전하여 10개 생성<br/>\n",
    "\n",
    "RGB pixel에 PCA를 하고, Gaussian dist(0, 0.1)을 곱하고 원래 픽셀에 더함"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "import os\n",
    "import cv2\n",
    "import random\n",
    "import matplotlib.pyplot as plt\n",
    "from matplotlib.mlab import PCA\n",
    "\n",
    "# path = \"/home/heim/mount_disk/workspace/datasets/101_ObjectCategories/\"\n",
    "path = \"/home/heim/mount_disk/workspace/datasets/ilsvrc12/train/\"\n",
    "\n",
    "# folders_train = os.listdir(path+\"train/\")\n",
    "# path_data = []\n",
    "# for i in range(folders_train.__len__()):\n",
    "#     data_aclass = os.listdir(path+\"train/\"+folders_train[i])\n",
    "#     path_data.append(data_aclass)\n",
    "\n",
    "def gether_data(batch_size):\n",
    "    x_batch = []\n",
    "    y_batch = []\n",
    "    \n",
    "    for i in range(batch_size):\n",
    "        y = np.random.randint(0,1000)\n",
    "        data_path = path+str(os.listdir(path)[y])+\"/\"\n",
    "        data_path +=random.choice(os.listdir(data_path))\n",
    "        im = cv2.imread(data_path).astype(float)\n",
    "        # im = cv2.cvtColor(im, cv2.COLOR_BGR2RGB)\n",
    "        t = int(im.shape[0]>im.shape[1])\n",
    "        diff = im.shape[1]-im.shape[0]\n",
    "        im = im[-t*int(diff/2):im.shape[0]+t*int(diff/2),(1-t)*int(diff/2):im.shape[1]-(1-t)*int(diff/2),:]\n",
    "        im = cv2.resize(im, (256,256))\n",
    "        \n",
    "        pos_x, pos_y = [[0,0],[2,2],[1,1],[0,2],[2,0]][np.random.randint(0,5)]\n",
    "        \n",
    "        im = im[16*pos_x:256-16*(2-pos_x),16*pos_y:256-16*(2-pos_y),:]\n",
    "        \n",
    "        if np.random.randint(0,2):\n",
    "            im = cv2.flip(im, 1)\n",
    "        # plt.imshow(im)\n",
    "        # plt.show()\n",
    "        \n",
    "        im =(im-127)/255\n",
    "        x_batch.append(im)\n",
    "        y_batch.append(y)\n",
    "    \n",
    "    return np.array(x_batch).transpose(0,3,1,2), one_hot_encoder(np.array(y_batch),1000)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(2, 3, 224, 224)"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "gether_data(2)[0].shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(2, 3, 224, 224) (2, 1000)\n"
     ]
    }
   ],
   "source": [
    "x_batch, y_batch = gether_data(2)\n",
    "print(x_batch.shape,y_batch.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Network structure\n",
    "Total 8 weight layers\n",
    "\n",
    "**5 Convolution layers**\n",
    "\n",
    "        Conv 1 : 11x11x3, 96 kernels, stride=4, pad=0? 2\n",
    "            ReLU\n",
    "            LRN\n",
    "            MaxPool 1 : 3x3, stride=2\n",
    "        Conv 2 : 5x5x96, 256 kernels, stride=1, pad=2\n",
    "            ReLU\n",
    "            LRN\n",
    "            MaxPool 2 : 3x3, stride=2\n",
    "        Conv 3 : 3x3x256, 384 kernels, stride=1, pad=1\n",
    "            ReLU\n",
    "        Conv 4 : 3x3x384, 384 kernels, stride=1, pad=1\n",
    "            ReLU\n",
    "        Conv 5 : 3x3x384, 256 kernels, stride=1, pad=1\n",
    "            ReLU\n",
    "            MaxPool 3 : 3x3, stride=2\n",
    "**3 Dense layers**\n",
    "\n",
    "        Dense 1:  4096\n",
    "            ReLU\n",
    "            Dropout : 50 %\n",
    "        Dense 2 : 4096\n",
    "            ReLU\n",
    "            Dropout : 50 %\n",
    "        Dense 3 : 1000\n",
    "    \n",
    "**1000-way softmax**\n",
    "\n",
    "![Alt text](http://csxcs366.eefocus.com/data/15-09/201042223040939/1442997038_30fbaec7.png)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 Convolution (1, 96, 55, 55)\n",
      "1 Relu (1, 96, 55, 55)\n",
      "2 Pooling (1, 96, 27, 27)\n",
      "3 Convolution (1, 256, 27, 27)\n",
      "4 Relu (1, 256, 27, 27)\n",
      "5 Pooling (1, 256, 13, 13)\n",
      "6 Convolution (1, 384, 13, 13)\n",
      "7 Relu (1, 384, 13, 13)\n",
      "8 Convolution (1, 384, 13, 13)\n",
      "9 Relu (1, 384, 13, 13)\n",
      "10 Convolution (1, 256, 13, 13)\n",
      "11 Relu (1, 256, 13, 13)\n",
      "12 Pooling (1, 256, 6, 6)\n",
      "13 Flatten (1, 9216)\n",
      "14 Affine (1, 4096)\n",
      "15 Relu (1, 4096)\n",
      "16 Affine (1, 4096)\n",
      "17 Relu (1, 4096)\n",
      "18 Affine (1, 1000)\n"
     ]
    }
   ],
   "source": [
    "net = Network()\n",
    "\n",
    "initializer =\"adam\"\n",
    "# Conv 1 : 11x11x3, 96 kernels, stride=4, pad=0\n",
    "net.add(Convolution([3,224,224],[11,11],96, stride=4, pad=2, initializer=initializer))\n",
    "# net.add(Convolution([3,224,224],[7,7],96, stride=2, pad=2, initializer=initializer))\n",
    "net.add(Relu())\n",
    "net.add(Pooling([3,3], stride=2))\n",
    "\n",
    "# Conv 2 : 5x5x96, 256 kernels, stride=1, pad=2\n",
    "net.add(Convolution([96,27,27],[5,5],256, pad=2, initializer=initializer))\n",
    "net.add(Relu())\n",
    "net.add(Pooling([3,3], stride=2))\n",
    "\n",
    "# Conv 3 : 3x3x256, 384 kernels, stride=1, pad=1\n",
    "net.add(Convolution([256,13,13],[5,5],384, pad=2, initializer=initializer))\n",
    "net.add(Relu())\n",
    "\n",
    "# net.add(Pooling([3,3], stride=2))\n",
    "\n",
    "# Conv 4 : 3x3x384, 384 kernels, stride=1, pad=1\n",
    "net.add(Convolution([384,13,13],[3,3],384, pad=1, initializer=initializer))\n",
    "net.add(Relu())\n",
    "\n",
    "# Conv 5 : 3x3x384, 256 kernels, stride=1, pad=1\n",
    "net.add(Convolution([384,13,13],[3,3],256, pad=1, initializer=initializer))\n",
    "net.add(Relu())\n",
    "net.add(Pooling([3,3], stride=2))\n",
    "\n",
    "\n",
    "net.add(Flatten())\n",
    "net.add(Affine([9216,4096], initializer=initializer))\n",
    "net.add(Relu())\n",
    "# net.add(Dropout())\n",
    "net.add(Affine([4096,4096], initializer=initializer))\n",
    "net.add(Relu())\n",
    "# net.add(Dropout())\n",
    "net.add(Affine([4096,1000], initializer=initializer))\n",
    "net.set_loss(SoftmaxWithLoss())\n",
    "\n",
    "i=0\n",
    "out = np.ones([1,3,224,224])\n",
    "for layer in net.layers:\n",
    "    out = layer.forward(out)\n",
    "    print(i , layer.__class__.__name__, out.shape)\n",
    "    i+=1"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Details of Learning\n",
    "- mini-batch size : 128\n",
    "- momentum : 0.9, weight decay : 0.0005\n",
    "- 0-mean gaussian *0.01\n",
    "- dropout:0.5\n",
    "- learning rate : 0.01, validation error가 멈추면 1/10배\n",
    "- (7개 CNN 앙상블) 미정"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch_accuracy_track = []\n",
    "test_accuracy_track = []\n",
    "loss_track=[]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "== Loss 5.9377, Process : 0.1995% ==\r"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-26-68d9a6d7cf0a>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m     10\u001b[0m     \u001b[0;32mfor\u001b[0m \u001b[0mi\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mrange\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0miter_step\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     11\u001b[0m         \u001b[0mx_batch\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_batch\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mgether_data\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mbatch_size\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 12\u001b[0;31m         \u001b[0mloss\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnet\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfit\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx_batch\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_batch\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlearning_rate\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mlearning_rate\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     13\u001b[0m         \u001b[0mloss_track\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mappend\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mloss\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     14\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mi\u001b[0m\u001b[0;34m%\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmax\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0miter_step\u001b[0m\u001b[0;34m/\u001b[0m\u001b[0;36m100000\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m==\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/mount_disk/workspace/python/jupyter/Deep learning from scratch/network.py\u001b[0m in \u001b[0;36mfit\u001b[0;34m(self, x, y, learning_rate, momentum_rate, adaGrad_rate, RMSProp_rate, optimizer)\u001b[0m\n\u001b[1;32m    316\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    317\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0mfit\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mx\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlearning_rate\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m0.01\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmomentum_rate\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0madaGrad_rate\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mRMSProp_rate\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0moptimizer\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m\"adam\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 318\u001b[0;31m         \u001b[0mout\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpredict\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtrain_flg\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mTrue\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    319\u001b[0m         \u001b[0mloss\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mloss_layer\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mforward\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mout\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    320\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/mount_disk/workspace/python/jupyter/Deep learning from scratch/network.py\u001b[0m in \u001b[0;36mpredict\u001b[0;34m(self, x, train_flg)\u001b[0m\n\u001b[1;32m    309\u001b[0m                 \u001b[0mout\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mlayer\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mforward\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mout\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtrain_flg\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    310\u001b[0m             \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 311\u001b[0;31m                 \u001b[0mout\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mlayer\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mforward\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mout\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    312\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0mout\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    313\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/mount_disk/workspace/python/jupyter/Deep learning from scratch/network.py\u001b[0m in \u001b[0;36mforward\u001b[0;34m(self, x)\u001b[0m\n\u001b[1;32m    233\u001b[0m             \u001b[0mint\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m+\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mshape\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m3\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0mfilter_w\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m/\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mstride\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    234\u001b[0m             ]\n\u001b[0;32m--> 235\u001b[0;31m         \u001b[0mout_c\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mim2col\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfilter_h\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfilter_w\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mstride\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpad\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    236\u001b[0m         \u001b[0mout_c\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mout_c\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mreshape\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfilter_h\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0mfilter_w\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    237\u001b[0m         \u001b[0marg_max\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0margmax\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mout_c\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0maxis\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/mount_disk/workspace/python/jupyter/Deep learning from scratch/myfunctions.py\u001b[0m in \u001b[0;36mim2col\u001b[0;34m(input_data, filter_h, filter_w, stride, pad)\u001b[0m\n\u001b[1;32m     30\u001b[0m             \u001b[0mcol\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m:\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mx\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m:\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m:\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mimg\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m:\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0my_max\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0mstride\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mx\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0mx_max\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0mstride\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     31\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 32\u001b[0;31m     \u001b[0mcol\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mcol\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtranspose\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m4\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m5\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m1\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m2\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m3\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mreshape\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mN\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0mout_h\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0mout_w\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m-\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     33\u001b[0m     \u001b[0;32mreturn\u001b[0m \u001b[0mcol\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     34\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "import time\n",
    "train_size = 12000000\n",
    "batch_size = 128\n",
    "learning_rate=0.00003\n",
    "iter_per_epoch = train_size/batch_size\n",
    "epochs = 1\n",
    "iter_step = int(iter_per_epoch)\n",
    "for epoch in range(epochs):\n",
    "    start_time = time.time()\n",
    "    for i in range(iter_step):\n",
    "        x_batch, y_batch = gether_data(batch_size)\n",
    "        loss = net.fit(x_batch, y_batch, learning_rate=learning_rate)\n",
    "        loss_track.append(loss)\n",
    "        if i%(np.max([iter_step/100000,1]))==0:\n",
    "            print(\"== Loss {0:0.4f}, Process : {1:0.4f}% ==\".format(loss, i/iter_step*100), end=\"\\r\")\n",
    "    batch_accuracy = net.accuracy(x_batch, y_batch)\n",
    "    x_test, y_test = gether_data(batch_size)\n",
    "    test_accuracy = net.accuracy(x_test, y_test)\n",
    "    batch_accuracy_track.append(batch_accuracy)\n",
    "    test_accuracy_track.append(test_accuracy)\n",
    "    print(\"== {0} Epoch, Loss {1:0.4f}, Batch Accuracy  {2:0.4f}, Test Accuracy : {3:0.3f} == ({4:0.2f} m)\".format(epoch+1, loss, batch_accuracy, test_accuracy, (time.time()-start_time)/60))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_test, y_test = gether_data(100)\n",
    "net.accuracy(x_test,y_test,top=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "l_track = np.array(loss_track)\n",
    "x = np.arange(l_track.size)\n",
    "plt.plot(x, l_track, label=\"loss\")\n",
    "# plt.ylim([0,8])\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f1cb0a1e358>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "b_track = np.array(batch_accuracy_track)\n",
    "t_track = np.array(test_accuracy_track)\n",
    "x = np.arange(b_track.size)\n",
    "plt.plot(x, b_track, label=\"batch\")\n",
    "plt.plot(x, t_track, label=\"test\")\n",
    "# plt.ylim([0,1])\n",
    "plt.legend(loc=2)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "###  SAVE ###\n",
    "net.accuracy(x_batch[0:1], y_batch[0:1])\n",
    "net.set_save_mode()\n",
    "dump = (net.layers, batch_accuracy_track, test_accuracy_track, loss_track)\n",
    "with open('alexnet.wb', 'wb') as f:\n",
    "    pickle.dump(dump, f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "###  LOAD ###\n",
    "with open('alexnet.wb', 'rb') as f:\n",
    "    dump = pickle.load(f)\n",
    "(net.layers, batch_accuracy_track, test_accuracy_track, loss_track) = dump"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 96 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "filter_show(net.layers[0].W)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.01021050333597761"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "net.layers[0].W[0,0,0,0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.010683897555013736"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "net.layers[0].dW[0,0,0,0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "5.8261850147029985"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "net.fit(x_batch, y_batch, learning_rate=learning_rate)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.01019659063671522"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "net.layers[0].W[0,0,0,0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.010683897555013736"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "net.layers[0].dW[0,0,0,0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "-3.416734245537036e-05"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "0.007187844113708725-0.007189976661670202"
   ]
  },
  {
   "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": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}