keras/regularization/mlp-mnist-l2.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## MLP MNIST with L2 regularizer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_7\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_21 (Dense)             (None, 256)               200960    \n",
      "_________________________________________________________________\n",
      "activation_21 (Activation)   (None, 256)               0         \n",
      "_________________________________________________________________\n",
      "dense_22 (Dense)             (None, 256)               65792     \n",
      "_________________________________________________________________\n",
      "activation_22 (Activation)   (None, 256)               0         \n",
      "_________________________________________________________________\n",
      "dense_23 (Dense)             (None, 10)                2570      \n",
      "_________________________________________________________________\n",
      "activation_23 (Activation)   (None, 10)                0         \n",
      "=================================================================\n",
      "Total params: 269,322\n",
      "Trainable params: 269,322\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "Train on 60000 samples\n",
      "Epoch 1/20\n",
      "60000/60000 [==============================] - 4s 70us/sample - loss: 1.2198 - accuracy: 0.7331\n",
      "Epoch 2/20\n",
      "60000/60000 [==============================] - 3s 58us/sample - loss: 0.5291 - accuracy: 0.8793\n",
      "Epoch 3/20\n",
      "60000/60000 [==============================] - 3s 51us/sample - loss: 0.4349 - accuracy: 0.8981\n",
      "Epoch 4/20\n",
      "60000/60000 [==============================] - 3s 51us/sample - loss: 0.3943 - accuracy: 0.9078\n",
      "Epoch 5/20\n",
      "60000/60000 [==============================] - 3s 53us/sample - loss: 0.3688 - accuracy: 0.9144\n",
      "Epoch 6/20\n",
      "60000/60000 [==============================] - 4s 59us/sample - loss: 0.3495 - accuracy: 0.9197\n",
      "Epoch 7/20\n",
      "60000/60000 [==============================] - 3s 53us/sample - loss: 0.3337 - accuracy: 0.9245\n",
      "Epoch 8/20\n",
      "60000/60000 [==============================] - 3s 53us/sample - loss: 0.3205 - accuracy: 0.9277\n",
      "Epoch 9/20\n",
      "60000/60000 [==============================] - 3s 53us/sample - loss: 0.3083 - accuracy: 0.9312\n",
      "Epoch 10/20\n",
      "60000/60000 [==============================] - 4s 63us/sample - loss: 0.2979 - accuracy: 0.9344\n",
      "Epoch 11/20\n",
      "60000/60000 [==============================] - 4s 62us/sample - loss: 0.2879 - accuracy: 0.9374\n",
      "Epoch 12/20\n",
      "60000/60000 [==============================] - 4s 62us/sample - loss: 0.2792 - accuracy: 0.9403\n",
      "Epoch 13/20\n",
      "60000/60000 [==============================] - 4s 63us/sample - loss: 0.2708 - accuracy: 0.9425\n",
      "Epoch 14/20\n",
      "60000/60000 [==============================] - 4s 59us/sample - loss: 0.2631 - accuracy: 0.9447\n",
      "Epoch 15/20\n",
      "60000/60000 [==============================] - 3s 54us/sample - loss: 0.2556 - accuracy: 0.9465\n",
      "Epoch 16/20\n",
      "60000/60000 [==============================] - 3s 55us/sample - loss: 0.2491 - accuracy: 0.9482\n",
      "Epoch 17/20\n",
      "60000/60000 [==============================] - 3s 54us/sample - loss: 0.2427 - accuracy: 0.9499\n",
      "Epoch 18/20\n",
      "60000/60000 [==============================] - 3s 55us/sample - loss: 0.2365 - accuracy: 0.9512\n",
      "Epoch 19/20\n",
      "60000/60000 [==============================] - 3s 58us/sample - loss: 0.2311 - accuracy: 0.9531\n",
      "Epoch 20/20\n",
      "60000/60000 [==============================] - 4s 65us/sample - loss: 0.2255 - accuracy: 0.9546\n",
      "\n",
      "Test accuracy: 95.3%\n"
     ]
    }
   ],
   "source": [
    "'''\n",
    "MLP network for MNIST digits classification with L2 reg\n",
    "Test accuracy: 95.3%\n",
    "'''\n",
    "\n",
    "from __future__ import absolute_import\n",
    "from __future__ import division\n",
    "from __future__ import print_function\n",
    "\n",
    "# numpy package\n",
    "import numpy as np\n",
    "from tensorflow.keras.models import Sequential\n",
    "from tensorflow.keras.layers import Dense, Activation\n",
    "from tensorflow.keras.datasets import mnist\n",
    "from tensorflow.keras.regularizers import l2, l1\n",
    "from tensorflow.keras.utils import to_categorical\n",
    "\n",
    "from numpy.random import seed\n",
    "seed(12345)\n",
    "import tensorflow as tf\n",
    "tf.random.set_seed(12345)\n",
    "\n",
    "# load mnist dataset\n",
    "(x_train, y_train), (x_test, y_test) = mnist.load_data()\n",
    "\n",
    "# compute the number of labels\n",
    "num_labels = len(np.unique(y_train))\n",
    "\n",
    "# convert to one-hot vector\n",
    "y_train = to_categorical(y_train)\n",
    "y_test = to_categorical(y_test)\n",
    "\n",
    "# image dimensions (assumed square)\n",
    "image_size = x_train.shape[1]\n",
    "input_size = image_size * image_size\n",
    "# for mlp, the input dim is a vector, so we reshape\n",
    "x_train = np.reshape(x_train, [-1, input_size])\n",
    "# we train our network using float data\n",
    "x_train = x_train.astype('float32') / 255\n",
    "x_test = np.reshape(x_test, [-1, input_size])\n",
    "x_test = x_test.astype('float32') / 255\n",
    "\n",
    "# network parameters\n",
    "batch_size = 128\n",
    "hidden_units = 256\n",
    "\n",
    "kernel_regularizer = l2(0.0001)\n",
    "# this is 3-layer MLP with ReLU and l2 kernel regularizer\n",
    "model = Sequential()\n",
    "model.add(Dense(hidden_units,\n",
    "                kernel_regularizer=kernel_regularizer,\n",
    "                input_dim=input_size))\n",
    "model.add(Activation('relu'))\n",
    "model.add(Dense(hidden_units,\n",
    "                kernel_regularizer=kernel_regularizer))\n",
    "model.add(Activation('relu'))\n",
    "model.add(Dense(num_labels))\n",
    "# this is the output for one-hot vector\n",
    "model.add(Activation('softmax'))\n",
    "model.summary()\n",
    "\n",
    "# loss function for one-hot vector\n",
    "# use of sgd optimizer\n",
    "# accuracy is good metric for classification tasks\n",
    "model.compile(loss='categorical_crossentropy',\n",
    "              optimizer='sgd',\n",
    "              metrics=['accuracy'])\n",
    "# train the network\n",
    "model.fit(x_train, y_train, epochs=20, batch_size=batch_size)\n",
    "\n",
    "# validate the model on test dataset to determine generalization\n",
    "score = model.evaluate(x_test,\n",
    "                       y_test,\n",
    "                       batch_size=batch_size,\n",
    "                       verbose=False)\n",
    "print(\"\\nTest accuracy: %.1f%%\" % (100.0 * score[1]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "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": "3.7.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}