conv_nocolors.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "dim_x, dim_y, colors=30,30,1\n",
    "kernel_x, kernel_y, in_colors, out_colors=3,3,colors, 4"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "im=np.random.rand(dim_x, dim_y, colors)\n",
    "weights=np.random.rand(kernel_x, kernel_y, colors)\n",
    "\n",
    "debug_weights=np.random.rand(*(dim_x-kernel_x+1, dim_y-kernel_y+1, colors))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(30, 30, 1)"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "im.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "def forward(im, weights):\n",
    "    kernel_x, kernel_y, colors=weights.shape\n",
    "    dim_x, dim_y, colors=im.shape\n",
    "    \n",
    "    \n",
    "    out=np.empty((dim_x-kernel_x+1, dim_y-kernel_y+1, colors))\n",
    "    \n",
    "    for i in range(out.shape[0]):\n",
    "        for j in range(out.shape[1]):\n",
    "            crop=im[i:i+kernel_x, j:j+kernel_y]\n",
    "            out[i,j]=np.sum(crop*weights)\n",
    "            \n",
    "    return out"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "out=forward(im, weights)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "debug_out=np.sum(out*debug_weights)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "889.11561753070669"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "debug_out"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "debug_shape=debug_weights.shape\n",
    "\n",
    "padded=np.zeros((debug_shape[0]+2*(kernel_x-1), debug_shape[1]+2*(kernel_y-1), colors))\n",
    "padded[kernel_x-1:kernel_x-1+debug_shape[0], kernel_y-1:kernel_y-1+debug_shape[1]]=debug_weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7f007c243978>"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP8AAAD8CAYAAAC4nHJkAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nMn6pKeOXmjJ+qSnjl5r6X8ImFDmY0vH+AAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f007e5360f0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow((padded>0).sum(axis=-1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "backward_weights=weights[::-1,::-1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "grads=forward(padded, backward_weights)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(30, 30, 1)"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "grads.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "eps=1e-4\n",
    "\n",
    "d_im=im.copy()\n",
    "d_im[4,4]+=eps"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "d_out=forward(d_im, weights)\n",
    "d_debug_out=np.sum(d_out*debug_weights)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "grad=(d_debug_out-debug_out)/eps"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1.7115924799782078"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "grad"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ 1.71159248])"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "grads[4,4]"
   ]
  },
  {
   "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": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}