demo_mnist.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import os\n",
    "import urllib\n",
    "import gzip\n",
    "import struct\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def download_data(url, force_download=False): \n",
    "    fname = url.split(\"/\")[-1]\n",
    "    if force_download or not os.path.exists(fname):\n",
    "        urllib.request.urlretrieve(url, fname)\n",
    "    return fname\n",
    "\n",
    "def read_data(label_url, image_url):\n",
    "    with gzip.open(download_data(label_url)) as flbl:\n",
    "        magic, num = struct.unpack(\">II\", flbl.read(8))\n",
    "        label = np.fromstring(flbl.read(), dtype=np.int8)\n",
    "    with gzip.open(download_data(image_url), 'rb') as fimg:\n",
    "        magic, num, rows, cols = struct.unpack(\">IIII\", fimg.read(16))\n",
    "        image = np.fromstring(fimg.read(), dtype=np.uint8).reshape(len(label), rows, cols)\n",
    "    return (label, image)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "path = 'http://yann.lecun.com/exdb/mnist/'\n",
    "\n",
    "(train_lbl, train_img) = read_data(\n",
    "    path+'train-labels-idx1-ubyte.gz', path+'train-images-idx3-ubyte.gz')\n",
    "\n",
    "(test_lbl, test_img) = read_data(\n",
    "    path+'t10k-labels-idx1-ubyte.gz', path+'t10k-images-idx3-ubyte.gz')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x227f77ea278>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "label: [5 0 4 1 9 2 1 3 1 4]\n"
     ]
    }
   ],
   "source": [
    "for i in range(10):\n",
    "    plt.subplot(1,10,i+1)\n",
    "    plt.imshow(train_img[i], cmap='Greys_r')\n",
    "    plt.axis('off')\n",
    "plt.show()\n",
    "print('label: %s' % (train_lbl[0:10],))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x227f77d7908>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(test_img[0], cmap='Greys_r')\n",
    "plt.axis('off')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "num_examples = 1000"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((1000, 28, 28), (1000,))"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_img = train_img[0:num_examples]\n",
    "train_lbl = train_lbl[0:num_examples]\n",
    "train_img.shape, train_lbl.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((10000, 28, 28), (10000,))"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_img.shape, test_lbl.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1000, 784)"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x_data = train_img.reshape(-1, 28 * 28) / np.float32(256)\n",
    "x_data.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1000, 10)"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def to_one_hot(labels, num_classes=10):\n",
    "    return np.eye(num_classes)[labels]\n",
    "\n",
    "y_labels = to_one_hot(train_lbl)\n",
    "y_labels.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "from layers.linear_layer import *\n",
    "from nets.fcnn import *"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "layers = [\n",
    "    LinearLayer(name=\"input\", n_in=x_data.shape[1], n_out=10, activation=sigmoid_function),\n",
    "    LinearLayer(name=\"hidden 1\", n_in=10, n_out=15, activation=sigmoid_function),\n",
    "    LinearLayer(name=\"hidden 2\", n_in=15, n_out=10, activation=sigmoid_function),\n",
    "    LinearLayer(name=\"output\", n_in=10, n_out=y_labels.shape[1], activation=sigmoid_function)\n",
    "]\n",
    "\n",
    "nn = FCNeuralNet(layers=layers, cost_func=subtract_err)\n",
    "# print_network(nn)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "network:\n",
      "layer - input: weights: (784, 10)\n",
      "layer - hidden 1: weights: (10, 15)\n",
      "layer - hidden 2: weights: (15, 10)\n",
      "layer - output: weights: (10, 10)\n"
     ]
    }
   ],
   "source": [
    "def print_network(nn):\n",
    "    print(\"network:\")\n",
    "    for layer in nn.layers:\n",
    "        print(\"layer - %s: weights: %s\" %\n",
    "              (layer.name, layer.weights.shape))\n",
    "        \n",
    "print_network(nn)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epochs: 0 error: 0.493481336173\n",
      "epochs: 500 error: 0.0374563272437\n",
      "epochs: 1000 error: 0.0181991110755\n",
      "epochs: 1500 error: 0.0135115834405\n",
      "epochs: 2000 error: 0.0113773689944\n",
      "epochs: 2500 error: 0.0098725520874\n",
      "epochs: 3000 error: 0.00891364673286\n",
      "epochs: 3500 error: 0.00826437578711\n",
      "epochs: 4000 error: 0.00779947141318\n",
      "epochs: 4500 error: 0.00760078586273\n",
      "epochs: 5000 error: 0.00741089182245\n",
      "epochs: 5500 error: 0.00702923873948\n",
      "epochs: 6000 error: 0.00670342965253\n",
      "epochs: 6500 error: 0.00646597127418\n",
      "epochs: 7000 error: 0.00616634963057\n",
      "epochs: 7500 error: 0.00589285366121\n",
      "epochs: 8000 error: 0.00573819200548\n",
      "epochs: 8500 error: 0.00569026592285\n",
      "epochs: 9000 error: 0.00548538589395\n",
      "epochs: 9500 error: 0.0053251899853\n"
     ]
    }
   ],
   "source": [
    "EPOCHS = 10000\n",
    "nn.train(x_data, y_labels, learning_rate=0.01, epochs=EPOCHS, batch_size=1, print_epochs=500)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(10000, 784)"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_data = (test_img.reshape(-1, 28 * 28) / np.float32(256))\n",
    "test_data.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "76.759999999999991"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_predictions = np.argmax(nn.predict(test_data), axis=1)\n",
    "acc = np.mean(test_predictions == test_lbl)\n",
    "acc * 100"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([7, 3, 1, ..., 9, 8, 6], dtype=int64)"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([7, 2, 1, ..., 4, 5, 6], dtype=int8)"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_lbl"
   ]
  },
  {
   "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": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}