CNN_training_example.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 89,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "from matplotlib.pyplot import imshow\n",
    "import numpy as np\n",
    "from PIL import Image, ImageDraw\n",
    "from keras import backend as K\n",
    "import random as rnd\n",
    "\n",
    "def makeRandomLabel():\n",
    "    labelSize = 25\n",
    "    bits = rnd.getrandbits(labelSize)\n",
    "    bitArray = np.zeros(labelSize)\n",
    "    for i in range(labelSize):\n",
    "        bitArray[i] = (bits & 0b1)        \n",
    "        bits = bits >> 1\n",
    "    return bitArray\n",
    "\n",
    "def drawImage(bitArray, imageSize):\n",
    "  \n",
    "    img = Image.new(\"RGB\", (imageSize, imageSize), \"white\")\n",
    "\n",
    "    draw = ImageDraw.Draw(img)\n",
    "    size = imageSize/30\n",
    "    offset = (rnd.randrange(int(imageSize/6), int(imageSize/2)), rnd.randrange(int(imageSize/6), int(imageSize/2)))\n",
    "    spacing = size + imageSize/37\n",
    "    numLights = 5\n",
    "\n",
    "    box = (offset[0]-spacing*1.5, offset[1]-spacing*1.5, offset[0]+spacing*6, offset[1]+spacing*6)\n",
    "    draw.ellipse(box, fill=0)\n",
    "\n",
    "    for y in range(numLights):\n",
    "        for x in range(numLights):\n",
    "            box = (offset[0] + x*spacing, offset[1] + y*spacing,offset[0] + x*spacing+size, offset[1] + y*spacing+size)\n",
    "            value = bitArray[x + y * numLights]\n",
    "            if value:\n",
    "                draw.ellipse(box, fill=\"white\")\n",
    "    del draw\n",
    "   \n",
    "    return (img)\n",
    "\n",
    "def traningSet(imageSize, batchSize=32):\n",
    "    while(True):\n",
    "        labels = np.ndarray(shape=(batchSize, 25), dtype='float32', order='F')\n",
    "        images = np.ndarray(shape=(batchSize, 3, imageSize, imageSize), dtype='float64', order='F')\n",
    "        for i in range(0, batchSize):\n",
    "            label = makeRandomLabel()\n",
    "            image = drawImage(label, imageSize)\n",
    "            labels[i] =  np.asarray(label, dtype='float32')\n",
    "            \n",
    "            imageData = np.asarray(image, dtype='float64')\n",
    "            images[i] = imageData.reshape(3, imageSize, imageSize)\n",
    "\n",
    "        yield (images, labels)\n",
    "                "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ 1.,  1.,  0.,  1.,  0.,  0.,  1.,  0.,  1.,  1.,  1.,  0.,  1.,\n",
       "        1.,  1.,  1.,  1.,  0.,  1.,  0.,  0.,  1.,  1.,  1.,  0.])"
      ]
     },
     "execution_count": 98,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "label = makeRandomLabel()\n",
    "label"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAASwAAAEsCAIAAAD2HxkiAAAGaUlEQVR4nO3d227bRhRAUbno//+y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\n",
      "text/plain": [
       "<PIL.Image.Image image mode=RGB size=300x300 at 0x7F54CDC755F8>"
      ]
     },
     "execution_count": 99,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "drawImage(label, 300)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 91,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def ConvBlock(layers, model, filters):\n",
    "    for i in range(layers): \n",
    "        model.add(ZeroPadding2D((1,1)))\n",
    "        model.add(Convolution2D(filters, 3, 3, activation='relu'))\n",
    "    model.add(MaxPooling2D((2,2), strides=(2,2)))\n",
    "    \n",
    "def FCBlock(model):\n",
    "    model.add(Dense(4096, activation='relu'))\n",
    "    model.add(Dropout(0.5))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/1\n",
      "2/2 [==============================] - 0s - loss: 37.0219 - acc: 0.0000e+00\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x7f54c642f160>"
      ]
     },
     "execution_count": 93,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from numpy.random import random, permutation\n",
    "from scipy import misc, ndimage\n",
    "from scipy.ndimage.interpolation import zoom\n",
    "\n",
    "import keras\n",
    "from keras import backend as K\n",
    "from keras.utils.data_utils import get_file\n",
    "from keras.models import Sequential, Model\n",
    "from keras.layers.core import Flatten, Dense, Dropout, Lambda\n",
    "from keras.layers.convolutional import Convolution2D, MaxPooling2D, ZeroPadding2D\n",
    "from keras.optimizers import SGD, RMSprop\n",
    "from keras.preprocessing import image\n",
    "\n",
    "imageSize = 300\n",
    "model = Sequential()\n",
    "model.add(Lambda(lambda x: x/256.0, input_shape=(3, imageSize, imageSize)))\n",
    "ConvBlock(2, model, 64)\n",
    "ConvBlock(2, model, 128)\n",
    "ConvBlock(3, model, 256)\n",
    "ConvBlock(3, model, 512)\n",
    "ConvBlock(3, model, 512)\n",
    "model.add(Flatten())\n",
    "FCBlock(model)\n",
    "FCBlock(model)\n",
    "model.add(Dense(25, activation='softmax'))\n",
    "model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])\n",
    "\n",
    "generatedBatch = traningSet(batchSize=2, imageSize=imageSize)\n",
    "model.fit_generator(generatedBatch, samples_per_epoch=2, nb_epoch=1)"
   ]
  },
  {
   "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": "3.4.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}