Siamese Network.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### A sample implemention of Siamese Network using Keras: https://github.com/keras-team/keras/blob/master/examples/mnist_siamese.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_72 (InputLayer)        (None, 28, 28)            0         \n",
      "_________________________________________________________________\n",
      "flatten_25 (Flatten)         (None, 784)               0         \n",
      "_________________________________________________________________\n",
      "dense_63 (Dense)             (None, 128)               100480    \n",
      "_________________________________________________________________\n",
      "dropout_39 (Dropout)         (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_64 (Dense)             (None, 128)               16512     \n",
      "_________________________________________________________________\n",
      "dropout_40 (Dropout)         (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_65 (Dense)             (None, 128)               16512     \n",
      "=================================================================\n",
      "Total params: 133,504\n",
      "Trainable params: 133,504\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "Train on 108400 samples, validate on 17820 samples\n",
      "Epoch 1/20\n",
      "108400/108400 [==============================] - 23s 212us/step - loss: 0.0968 - accuracy: 0.8853 - val_loss: 0.0530 - val_accuracy: 0.9448\n",
      "Epoch 2/20\n",
      "108400/108400 [==============================] - 16s 149us/step - loss: 0.0406 - accuracy: 0.9596 - val_loss: 0.0355 - val_accuracy: 0.9636\n",
      "Epoch 3/20\n",
      "108400/108400 [==============================] - 17s 154us/step - loss: 0.0280 - accuracy: 0.9725 - val_loss: 0.0308 - val_accuracy: 0.9659\n",
      "Epoch 4/20\n",
      "108400/108400 [==============================] - 16s 150us/step - loss: 0.0224 - accuracy: 0.9781 - val_loss: 0.0291 - val_accuracy: 0.9669\n",
      "Epoch 5/20\n",
      "108400/108400 [==============================] - 16s 147us/step - loss: 0.0194 - accuracy: 0.9806 - val_loss: 0.0287 - val_accuracy: 0.9681\n",
      "Epoch 6/20\n",
      "108400/108400 [==============================] - 17s 156us/step - loss: 0.0171 - accuracy: 0.9827 - val_loss: 0.0280 - val_accuracy: 0.9687\n",
      "Epoch 7/20\n",
      "108400/108400 [==============================] - 16s 152us/step - loss: 0.0157 - accuracy: 0.9846 - val_loss: 0.0282 - val_accuracy: 0.9682\n",
      "Epoch 8/20\n",
      "108400/108400 [==============================] - 15s 141us/step - loss: 0.0142 - accuracy: 0.9859 - val_loss: 0.0279 - val_accuracy: 0.9692\n",
      "Epoch 9/20\n",
      "108400/108400 [==============================] - 17s 154us/step - loss: 0.0132 - accuracy: 0.9865 - val_loss: 0.0293 - val_accuracy: 0.9678\n",
      "Epoch 10/20\n",
      "108400/108400 [==============================] - 17s 156us/step - loss: 0.0125 - accuracy: 0.9875 - val_loss: 0.0274 - val_accuracy: 0.9696\n",
      "Epoch 11/20\n",
      "108400/108400 [==============================] - 17s 158us/step - loss: 0.0116 - accuracy: 0.9882 - val_loss: 0.0282 - val_accuracy: 0.9679\n",
      "Epoch 12/20\n",
      "108400/108400 [==============================] - 18s 162us/step - loss: 0.0110 - accuracy: 0.9891 - val_loss: 0.0283 - val_accuracy: 0.9674\n",
      "Epoch 13/20\n",
      "108400/108400 [==============================] - 15s 142us/step - loss: 0.0108 - accuracy: 0.9889 - val_loss: 0.0281 - val_accuracy: 0.9684\n",
      "Epoch 14/20\n",
      "108400/108400 [==============================] - 17s 161us/step - loss: 0.0104 - accuracy: 0.9895 - val_loss: 0.0286 - val_accuracy: 0.9668\n",
      "Epoch 15/20\n",
      "108400/108400 [==============================] - 19s 175us/step - loss: 0.0098 - accuracy: 0.9899 - val_loss: 0.0285 - val_accuracy: 0.9685\n",
      "Epoch 16/20\n",
      "108400/108400 [==============================] - 18s 164us/step - loss: 0.0097 - accuracy: 0.9903 - val_loss: 0.0278 - val_accuracy: 0.9686\n",
      "Epoch 17/20\n",
      "108400/108400 [==============================] - 18s 163us/step - loss: 0.0093 - accuracy: 0.9905 - val_loss: 0.0275 - val_accuracy: 0.9690\n",
      "Epoch 18/20\n",
      "108400/108400 [==============================] - 18s 166us/step - loss: 0.0090 - accuracy: 0.9909 - val_loss: 0.0279 - val_accuracy: 0.9685\n",
      "Epoch 19/20\n",
      "108400/108400 [==============================] - 18s 164us/step - loss: 0.0090 - accuracy: 0.9908 - val_loss: 0.0298 - val_accuracy: 0.9674\n",
      "Epoch 20/20\n",
      "108400/108400 [==============================] - 17s 161us/step - loss: 0.0087 - accuracy: 0.9912 - val_loss: 0.0294 - val_accuracy: 0.9679\n",
      "* Accuracy on training set: 99.11%\n",
      "* Accuracy on test set: 96.69%\n"
     ]
    }
   ],
   "source": [
    "'''Trains a Siamese MLP on pairs of digits from the MNIST dataset.\n",
    "It follows Hadsell-et-al.'06 [1] by computing the Euclidean distance on the\n",
    "output of the shared network and by optimizing the contrastive loss (see paper\n",
    "for mode details).\n",
    "# References\n",
    "- Dimensionality Reduction by Learning an Invariant Mapping\n",
    "    http://yann.lecun.com/exdb/publis/pdf/hadsell-chopra-lecun-06.pdf\n",
    "Gets to 97.2% test accuracy after 20 epochs.\n",
    "2 seconds per epoch on a Titan X Maxwell GPU\n",
    "'''\n",
    "from __future__ import absolute_import\n",
    "from __future__ import print_function\n",
    "import numpy as np\n",
    "\n",
    "import random\n",
    "from keras.datasets import mnist\n",
    "from keras.models import Model\n",
    "from keras.layers import Input, Flatten, Dense, Dropout, Lambda\n",
    "from keras.optimizers import RMSprop\n",
    "from keras import backend as K\n",
    "\n",
    "num_classes = 10\n",
    "epochs = 20\n",
    "\n",
    "\n",
    "def euclidean_distance(vects):\n",
    "    x, y = vects\n",
    "    sum_square = K.sum(K.square(x - y), axis=1, keepdims=True)\n",
    "    return K.sqrt(K.maximum(sum_square, K.epsilon()))\n",
    "\n",
    "\n",
    "def eucl_dist_output_shape(shapes):\n",
    "    shape1, shape2 = shapes\n",
    "    return (shape1[0], 1)\n",
    "\n",
    "\n",
    "def contrastive_loss(y_true, y_pred):\n",
    "    '''Contrastive loss from Hadsell-et-al.'06\n",
    "    http://yann.lecun.com/exdb/publis/pdf/hadsell-chopra-lecun-06.pdf\n",
    "    '''\n",
    "    margin = 1\n",
    "    sqaure_pred = K.square(y_pred)\n",
    "    margin_square = K.square(K.maximum(margin - y_pred, 0))\n",
    "    return K.mean(y_true * sqaure_pred + (1 - y_true) * margin_square)\n",
    "\n",
    "\n",
    "def create_pairs(x, digit_indices):\n",
    "    '''Positive and negative pair creation.\n",
    "    Alternates between positive and negative pairs.\n",
    "    '''\n",
    "    pairs = []\n",
    "    labels = []\n",
    "    n = min([len(digit_indices[d]) for d in range(num_classes)]) - 1\n",
    "    for d in range(num_classes):\n",
    "        for i in range(n):\n",
    "            z1, z2 = digit_indices[d][i], digit_indices[d][i + 1]\n",
    "            pairs += [[x[z1], x[z2]]]\n",
    "            inc = random.randrange(1, num_classes)\n",
    "            dn = (d + inc) % num_classes\n",
    "            z1, z2 = digit_indices[d][i], digit_indices[dn][i]\n",
    "            pairs += [[x[z1], x[z2]]]\n",
    "            labels += [1, 0]\n",
    "    return np.array(pairs), np.array(labels)\n",
    "\n",
    "\n",
    "def create_base_network(input_shape):\n",
    "    '''Base network to be shared (eq. to feature extraction).\n",
    "    '''\n",
    "    input = Input(shape=input_shape)\n",
    "    x = Flatten()(input)\n",
    "    x = Dense(128, activation='relu')(x)\n",
    "    x = Dropout(0.1)(x)\n",
    "    x = Dense(128, activation='relu')(x)\n",
    "    x = Dropout(0.1)(x)\n",
    "    x = Dense(128, activation='relu')(x)\n",
    "    return Model(input, x)\n",
    "\n",
    "\n",
    "def compute_accuracy(y_true, y_pred):\n",
    "    '''Compute classification accuracy with a fixed threshold on distances.\n",
    "    '''\n",
    "    pred = y_pred.ravel() < 0.5\n",
    "    return np.mean(pred == y_true)\n",
    "\n",
    "\n",
    "def accuracy(y_true, y_pred):\n",
    "    '''Compute classification accuracy with a fixed threshold on distances.\n",
    "    '''\n",
    "    return K.mean(K.equal(y_true, K.cast(y_pred < 0.5, y_true.dtype)))\n",
    "\n",
    "\n",
    "# the data, split between train and test sets\n",
    "(x_train, y_train), (x_test, y_test) = mnist.load_data()\n",
    "x_train = x_train.astype('float32')\n",
    "x_test = x_test.astype('float32')\n",
    "x_train /= 255\n",
    "x_test /= 255\n",
    "input_shape = x_train.shape[1:]\n",
    "\n",
    "# create training+test positive and negative pairs\n",
    "digit_indices = [np.where(y_train == i)[0] for i in range(num_classes)]\n",
    "tr_pairs, tr_y = create_pairs(x_train, digit_indices)\n",
    "\n",
    "digit_indices = [np.where(y_test == i)[0] for i in range(num_classes)]\n",
    "te_pairs, te_y = create_pairs(x_test, digit_indices)\n",
    "\n",
    "# network definition\n",
    "base_network = create_base_network(input_shape)\n",
    "\n",
    "base_network.summary()\n",
    "input_a = Input(shape=input_shape)\n",
    "input_b = Input(shape=input_shape)\n",
    "\n",
    "# because we re-use the same instance `base_network`,\n",
    "# the weights of the network\n",
    "# will be shared across the two branches\n",
    "processed_a = base_network(input_a)\n",
    "processed_b = base_network(input_b)\n",
    "\n",
    "distance = Lambda(euclidean_distance,\n",
    "                  output_shape=eucl_dist_output_shape)([processed_a, processed_b])\n",
    "\n",
    "model = Model([input_a, input_b], distance)\n",
    "\n",
    "# train\n",
    "rms = RMSprop()\n",
    "model.compile(loss=contrastive_loss, optimizer=rms, metrics=[accuracy])\n",
    "model.fit([tr_pairs[:, 0], tr_pairs[:, 1]], tr_y,\n",
    "          batch_size=128,\n",
    "          epochs=epochs,\n",
    "          validation_data=([te_pairs[:, 0], te_pairs[:, 1]], te_y))\n",
    "\n",
    "# compute final accuracy on training and test sets\n",
    "y_pred = model.predict([tr_pairs[:, 0], tr_pairs[:, 1]])\n",
    "tr_acc = compute_accuracy(tr_y, y_pred)\n",
    "y_pred = model.predict([te_pairs[:, 0], te_pairs[:, 1]])\n",
    "te_acc = compute_accuracy(te_y, y_pred)\n",
    "\n",
    "print('* Accuracy on training set: %0.2f%%' % (100 * tr_acc))\n",
    "print('* Accuracy on test set: %0.2f%%' % (100 * te_acc))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### A litter modification demostrates how to apply your own model. Be carefull your input_shape, especially the channel.\n",
    "### I used the ResNet50 implemention based on Andrew Ng's course on [Coursera](https://github.com/hktxt/Deep-Learning-Specialization/blob/master/Course%204%20Convolutional%20Neural%20Networks/assignment/4.%20Residual%2BNetworks%2B-%2Bv2.ipynb)\n",
    "### even that a is huge network, which is unsatisfied with such samll size(28*28), however, it only for show how to do this. check blew:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_78 (InputLayer)           (None, 28, 28, 1)    0                                            \n",
      "__________________________________________________________________________________________________\n",
      "zero_padding2d_7 (ZeroPadding2D (None, 34, 34, 1)    0           input_78[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "conv1 (Conv2D)                  (None, 14, 14, 64)   3200        zero_padding2d_7[0][0]           \n",
      "__________________________________________________________________________________________________\n",
      "bn_conv1 (BatchNormalization)   (None, 14, 14, 64)   256         conv1[0][0]                      \n",
      "__________________________________________________________________________________________________\n",
      "activation_295 (Activation)     (None, 14, 14, 64)   0           bn_conv1[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling2d_7 (MaxPooling2D)  (None, 6, 6, 64)     0           activation_295[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "res2a_branch2a (Conv2D)         (None, 6, 6, 64)     4160        max_pooling2d_7[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "bn2a_branch2a (BatchNormalizati (None, 6, 6, 64)     256         res2a_branch2a[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_296 (Activation)     (None, 6, 6, 64)     0           bn2a_branch2a[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res2a_branch2b (Conv2D)         (None, 6, 6, 64)     36928       activation_296[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn2a_branch2b (BatchNormalizati (None, 6, 6, 64)     256         res2a_branch2b[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_297 (Activation)     (None, 6, 6, 64)     0           bn2a_branch2b[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res2a_branch2c (Conv2D)         (None, 6, 6, 256)    16640       activation_297[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "res2a_branch1 (Conv2D)          (None, 6, 6, 256)    16640       max_pooling2d_7[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "bn2a_branch2c (BatchNormalizati (None, 6, 6, 256)    1024        res2a_branch2c[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn2a_branch1 (BatchNormalizatio (None, 6, 6, 256)    1024        res2a_branch1[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "add_97 (Add)                    (None, 6, 6, 256)    0           bn2a_branch2c[0][0]              \n",
      "                                                                 bn2a_branch1[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "activation_298 (Activation)     (None, 6, 6, 256)    0           add_97[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "res2b_branch2a (Conv2D)         (None, 6, 6, 64)     16448       activation_298[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn2b_branch2a (BatchNormalizati (None, 6, 6, 64)     256         res2b_branch2a[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_299 (Activation)     (None, 6, 6, 64)     0           bn2b_branch2a[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res2b_branch2b (Conv2D)         (None, 6, 6, 64)     36928       activation_299[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn2b_branch2b (BatchNormalizati (None, 6, 6, 64)     256         res2b_branch2b[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_300 (Activation)     (None, 6, 6, 64)     0           bn2b_branch2b[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res2b_branch2c (Conv2D)         (None, 6, 6, 256)    16640       activation_300[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn2b_branch2c (BatchNormalizati (None, 6, 6, 256)    1024        res2b_branch2c[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "add_98 (Add)                    (None, 6, 6, 256)    0           bn2b_branch2c[0][0]              \n",
      "                                                                 activation_298[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_301 (Activation)     (None, 6, 6, 256)    0           add_98[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "res2c_branch2a (Conv2D)         (None, 6, 6, 64)     16448       activation_301[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn2c_branch2a (BatchNormalizati (None, 6, 6, 64)     256         res2c_branch2a[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_302 (Activation)     (None, 6, 6, 64)     0           bn2c_branch2a[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res2c_branch2b (Conv2D)         (None, 6, 6, 64)     36928       activation_302[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn2c_branch2b (BatchNormalizati (None, 6, 6, 64)     256         res2c_branch2b[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_303 (Activation)     (None, 6, 6, 64)     0           bn2c_branch2b[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res2c_branch2c (Conv2D)         (None, 6, 6, 256)    16640       activation_303[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn2c_branch2c (BatchNormalizati (None, 6, 6, 256)    1024        res2c_branch2c[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "add_99 (Add)                    (None, 6, 6, 256)    0           bn2c_branch2c[0][0]              \n",
      "                                                                 activation_301[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_304 (Activation)     (None, 6, 6, 256)    0           add_99[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "res3a_branch2a (Conv2D)         (None, 3, 3, 128)    32896       activation_304[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn3a_branch2a (BatchNormalizati (None, 3, 3, 128)    512         res3a_branch2a[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_305 (Activation)     (None, 3, 3, 128)    0           bn3a_branch2a[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res3a_branch2b (Conv2D)         (None, 3, 3, 128)    147584      activation_305[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn3a_branch2b (BatchNormalizati (None, 3, 3, 128)    512         res3a_branch2b[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_306 (Activation)     (None, 3, 3, 128)    0           bn3a_branch2b[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res3a_branch2c (Conv2D)         (None, 3, 3, 512)    66048       activation_306[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "res3a_branch1 (Conv2D)          (None, 3, 3, 512)    131584      activation_304[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn3a_branch2c (BatchNormalizati (None, 3, 3, 512)    2048        res3a_branch2c[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn3a_branch1 (BatchNormalizatio (None, 3, 3, 512)    2048        res3a_branch1[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "add_100 (Add)                   (None, 3, 3, 512)    0           bn3a_branch2c[0][0]              \n",
      "                                                                 bn3a_branch1[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "activation_307 (Activation)     (None, 3, 3, 512)    0           add_100[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "res3b_branch2a (Conv2D)         (None, 3, 3, 128)    65664       activation_307[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn3b_branch2a (BatchNormalizati (None, 3, 3, 128)    512         res3b_branch2a[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_308 (Activation)     (None, 3, 3, 128)    0           bn3b_branch2a[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res3b_branch2b (Conv2D)         (None, 3, 3, 128)    147584      activation_308[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn3b_branch2b (BatchNormalizati (None, 3, 3, 128)    512         res3b_branch2b[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_309 (Activation)     (None, 3, 3, 128)    0           bn3b_branch2b[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res3b_branch2c (Conv2D)         (None, 3, 3, 512)    66048       activation_309[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn3b_branch2c (BatchNormalizati (None, 3, 3, 512)    2048        res3b_branch2c[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "add_101 (Add)                   (None, 3, 3, 512)    0           bn3b_branch2c[0][0]              \n",
      "                                                                 activation_307[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_310 (Activation)     (None, 3, 3, 512)    0           add_101[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "res3c_branch2a (Conv2D)         (None, 3, 3, 128)    65664       activation_310[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn3c_branch2a (BatchNormalizati (None, 3, 3, 128)    512         res3c_branch2a[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_311 (Activation)     (None, 3, 3, 128)    0           bn3c_branch2a[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res3c_branch2b (Conv2D)         (None, 3, 3, 128)    147584      activation_311[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn3c_branch2b (BatchNormalizati (None, 3, 3, 128)    512         res3c_branch2b[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_312 (Activation)     (None, 3, 3, 128)    0           bn3c_branch2b[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res3c_branch2c (Conv2D)         (None, 3, 3, 512)    66048       activation_312[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn3c_branch2c (BatchNormalizati (None, 3, 3, 512)    2048        res3c_branch2c[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "add_102 (Add)                   (None, 3, 3, 512)    0           bn3c_branch2c[0][0]              \n",
      "                                                                 activation_310[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_313 (Activation)     (None, 3, 3, 512)    0           add_102[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "res3d_branch2a (Conv2D)         (None, 3, 3, 128)    65664       activation_313[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn3d_branch2a (BatchNormalizati (None, 3, 3, 128)    512         res3d_branch2a[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_314 (Activation)     (None, 3, 3, 128)    0           bn3d_branch2a[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res3d_branch2b (Conv2D)         (None, 3, 3, 128)    147584      activation_314[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn3d_branch2b (BatchNormalizati (None, 3, 3, 128)    512         res3d_branch2b[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_315 (Activation)     (None, 3, 3, 128)    0           bn3d_branch2b[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res3d_branch2c (Conv2D)         (None, 3, 3, 512)    66048       activation_315[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn3d_branch2c (BatchNormalizati (None, 3, 3, 512)    2048        res3d_branch2c[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "add_103 (Add)                   (None, 3, 3, 512)    0           bn3d_branch2c[0][0]              \n",
      "                                                                 activation_313[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_316 (Activation)     (None, 3, 3, 512)    0           add_103[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "res4a_branch2a (Conv2D)         (None, 2, 2, 256)    131328      activation_316[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn4a_branch2a (BatchNormalizati (None, 2, 2, 256)    1024        res4a_branch2a[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_317 (Activation)     (None, 2, 2, 256)    0           bn4a_branch2a[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res4a_branch2b (Conv2D)         (None, 2, 2, 256)    590080      activation_317[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn4a_branch2b (BatchNormalizati (None, 2, 2, 256)    1024        res4a_branch2b[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_318 (Activation)     (None, 2, 2, 256)    0           bn4a_branch2b[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res4a_branch2c (Conv2D)         (None, 2, 2, 1024)   263168      activation_318[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "res4a_branch1 (Conv2D)          (None, 2, 2, 1024)   525312      activation_316[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn4a_branch2c (BatchNormalizati (None, 2, 2, 1024)   4096        res4a_branch2c[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn4a_branch1 (BatchNormalizatio (None, 2, 2, 1024)   4096        res4a_branch1[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "add_104 (Add)                   (None, 2, 2, 1024)   0           bn4a_branch2c[0][0]              \n",
      "                                                                 bn4a_branch1[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "activation_319 (Activation)     (None, 2, 2, 1024)   0           add_104[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "res4b_branch2a (Conv2D)         (None, 2, 2, 256)    262400      activation_319[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn4b_branch2a (BatchNormalizati (None, 2, 2, 256)    1024        res4b_branch2a[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_320 (Activation)     (None, 2, 2, 256)    0           bn4b_branch2a[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res4b_branch2b (Conv2D)         (None, 2, 2, 256)    590080      activation_320[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn4b_branch2b (BatchNormalizati (None, 2, 2, 256)    1024        res4b_branch2b[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_321 (Activation)     (None, 2, 2, 256)    0           bn4b_branch2b[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res4b_branch2c (Conv2D)         (None, 2, 2, 1024)   263168      activation_321[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn4b_branch2c (BatchNormalizati (None, 2, 2, 1024)   4096        res4b_branch2c[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "add_105 (Add)                   (None, 2, 2, 1024)   0           bn4b_branch2c[0][0]              \n",
      "                                                                 activation_319[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_322 (Activation)     (None, 2, 2, 1024)   0           add_105[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "res4c_branch2a (Conv2D)         (None, 2, 2, 256)    262400      activation_322[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn4c_branch2a (BatchNormalizati (None, 2, 2, 256)    1024        res4c_branch2a[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_323 (Activation)     (None, 2, 2, 256)    0           bn4c_branch2a[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res4c_branch2b (Conv2D)         (None, 2, 2, 256)    590080      activation_323[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn4c_branch2b (BatchNormalizati (None, 2, 2, 256)    1024        res4c_branch2b[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_324 (Activation)     (None, 2, 2, 256)    0           bn4c_branch2b[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res4c_branch2c (Conv2D)         (None, 2, 2, 1024)   263168      activation_324[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn4c_branch2c (BatchNormalizati (None, 2, 2, 1024)   4096        res4c_branch2c[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "add_106 (Add)                   (None, 2, 2, 1024)   0           bn4c_branch2c[0][0]              \n",
      "                                                                 activation_322[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_325 (Activation)     (None, 2, 2, 1024)   0           add_106[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "res4d_branch2a (Conv2D)         (None, 2, 2, 256)    262400      activation_325[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn4d_branch2a (BatchNormalizati (None, 2, 2, 256)    1024        res4d_branch2a[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_326 (Activation)     (None, 2, 2, 256)    0           bn4d_branch2a[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res4d_branch2b (Conv2D)         (None, 2, 2, 256)    590080      activation_326[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn4d_branch2b (BatchNormalizati (None, 2, 2, 256)    1024        res4d_branch2b[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_327 (Activation)     (None, 2, 2, 256)    0           bn4d_branch2b[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res4d_branch2c (Conv2D)         (None, 2, 2, 1024)   263168      activation_327[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn4d_branch2c (BatchNormalizati (None, 2, 2, 1024)   4096        res4d_branch2c[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "add_107 (Add)                   (None, 2, 2, 1024)   0           bn4d_branch2c[0][0]              \n",
      "                                                                 activation_325[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_328 (Activation)     (None, 2, 2, 1024)   0           add_107[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "res4e_branch2a (Conv2D)         (None, 2, 2, 256)    262400      activation_328[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn4e_branch2a (BatchNormalizati (None, 2, 2, 256)    1024        res4e_branch2a[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_329 (Activation)     (None, 2, 2, 256)    0           bn4e_branch2a[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res4e_branch2b (Conv2D)         (None, 2, 2, 256)    590080      activation_329[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn4e_branch2b (BatchNormalizati (None, 2, 2, 256)    1024        res4e_branch2b[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_330 (Activation)     (None, 2, 2, 256)    0           bn4e_branch2b[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res4e_branch2c (Conv2D)         (None, 2, 2, 1024)   263168      activation_330[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn4e_branch2c (BatchNormalizati (None, 2, 2, 1024)   4096        res4e_branch2c[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "add_108 (Add)                   (None, 2, 2, 1024)   0           bn4e_branch2c[0][0]              \n",
      "                                                                 activation_328[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_331 (Activation)     (None, 2, 2, 1024)   0           add_108[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "res4f_branch2a (Conv2D)         (None, 2, 2, 256)    262400      activation_331[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn4f_branch2a (BatchNormalizati (None, 2, 2, 256)    1024        res4f_branch2a[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_332 (Activation)     (None, 2, 2, 256)    0           bn4f_branch2a[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res4f_branch2b (Conv2D)         (None, 2, 2, 256)    590080      activation_332[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn4f_branch2b (BatchNormalizati (None, 2, 2, 256)    1024        res4f_branch2b[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_333 (Activation)     (None, 2, 2, 256)    0           bn4f_branch2b[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res4f_branch2c (Conv2D)         (None, 2, 2, 1024)   263168      activation_333[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn4f_branch2c (BatchNormalizati (None, 2, 2, 1024)   4096        res4f_branch2c[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "add_109 (Add)                   (None, 2, 2, 1024)   0           bn4f_branch2c[0][0]              \n",
      "                                                                 activation_331[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_334 (Activation)     (None, 2, 2, 1024)   0           add_109[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "res5a_branch2a (Conv2D)         (None, 1, 1, 512)    524800      activation_334[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn5a_branch2a (BatchNormalizati (None, 1, 1, 512)    2048        res5a_branch2a[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_335 (Activation)     (None, 1, 1, 512)    0           bn5a_branch2a[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res5a_branch2b (Conv2D)         (None, 1, 1, 512)    2359808     activation_335[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn5a_branch2b (BatchNormalizati (None, 1, 1, 512)    2048        res5a_branch2b[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_336 (Activation)     (None, 1, 1, 512)    0           bn5a_branch2b[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res5a_branch2c (Conv2D)         (None, 1, 1, 2048)   1050624     activation_336[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "res5a_branch1 (Conv2D)          (None, 1, 1, 2048)   2099200     activation_334[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn5a_branch2c (BatchNormalizati (None, 1, 1, 2048)   8192        res5a_branch2c[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn5a_branch1 (BatchNormalizatio (None, 1, 1, 2048)   8192        res5a_branch1[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "add_110 (Add)                   (None, 1, 1, 2048)   0           bn5a_branch2c[0][0]              \n",
      "                                                                 bn5a_branch1[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "activation_337 (Activation)     (None, 1, 1, 2048)   0           add_110[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "res5b_branch2a (Conv2D)         (None, 1, 1, 512)    1049088     activation_337[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn5b_branch2a (BatchNormalizati (None, 1, 1, 512)    2048        res5b_branch2a[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_338 (Activation)     (None, 1, 1, 512)    0           bn5b_branch2a[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res5b_branch2b (Conv2D)         (None, 1, 1, 512)    2359808     activation_338[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn5b_branch2b (BatchNormalizati (None, 1, 1, 512)    2048        res5b_branch2b[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_339 (Activation)     (None, 1, 1, 512)    0           bn5b_branch2b[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res5b_branch2c (Conv2D)         (None, 1, 1, 2048)   1050624     activation_339[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn5b_branch2c (BatchNormalizati (None, 1, 1, 2048)   8192        res5b_branch2c[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "add_111 (Add)                   (None, 1, 1, 2048)   0           bn5b_branch2c[0][0]              \n",
      "                                                                 activation_337[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_340 (Activation)     (None, 1, 1, 2048)   0           add_111[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "res5c_branch2a (Conv2D)         (None, 1, 1, 512)    1049088     activation_340[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn5c_branch2a (BatchNormalizati (None, 1, 1, 512)    2048        res5c_branch2a[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_341 (Activation)     (None, 1, 1, 512)    0           bn5c_branch2a[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res5c_branch2b (Conv2D)         (None, 1, 1, 512)    2359808     activation_341[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn5c_branch2b (BatchNormalizati (None, 1, 1, 512)    2048        res5c_branch2b[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_342 (Activation)     (None, 1, 1, 512)    0           bn5c_branch2b[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "res5c_branch2c (Conv2D)         (None, 1, 1, 2048)   1050624     activation_342[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "bn5c_branch2c (BatchNormalizati (None, 1, 1, 2048)   8192        res5c_branch2c[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "add_112 (Add)                   (None, 1, 1, 2048)   0           bn5c_branch2c[0][0]              \n",
      "                                                                 activation_340[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "activation_343 (Activation)     (None, 1, 1, 2048)   0           add_112[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "flatten_27 (Flatten)            (None, 2048)         0           activation_343[0][0]             \n",
      "__________________________________________________________________________________________________\n",
      "dense_67 (Dense)                (None, 128)          262272      flatten_27[0][0]                 \n",
      "==================================================================================================\n",
      "Total params: 23,843,712\n",
      "Trainable params: 23,790,592\n",
      "Non-trainable params: 53,120\n",
      "__________________________________________________________________________________________________\n",
      "Train on 108400 samples, validate on 17820 samples\n",
      "Epoch 1/1\n",
      "108400/108400 [==============================] - 137402s 1s/step - loss: 1.0951 - accuracy: 0.5001 - val_loss: 0.4997 - val_accuracy: 0.5000\n",
      "* Accuracy on training set: 50.00%\n",
      "* Accuracy on test set: 50.00%\n"
     ]
    }
   ],
   "source": [
    "from __future__ import absolute_import\n",
    "from __future__ import print_function\n",
    "import numpy as np\n",
    "\n",
    "import random\n",
    "from keras.datasets import mnist\n",
    "from keras.models import Model\n",
    "from keras.layers import Input, Flatten, Dense, Dropout, Lambda, Conv2D\n",
    "from keras.optimizers import RMSprop\n",
    "from keras import backend as K\n",
    "from Siam_model import *\n",
    "\n",
    "num_classes = 10\n",
    "epochs = 1\n",
    "\n",
    "\n",
    "def euclidean_distance(vects):\n",
    "    x, y = vects\n",
    "    sum_square = K.sum(K.square(x - y), axis=1, keepdims=True)\n",
    "    return K.sqrt(K.maximum(sum_square, K.epsilon()))\n",
    "\n",
    "\n",
    "def eucl_dist_output_shape(shapes):\n",
    "    shape1, shape2 = shapes\n",
    "    return (shape1[0], 1)\n",
    "\n",
    "\n",
    "def contrastive_loss(y_true, y_pred):\n",
    "\n",
    "    margin = 1\n",
    "    sqaure_pred = K.square(y_pred)\n",
    "    margin_square = K.square(K.maximum(margin - y_pred, 0))\n",
    "    return K.mean(y_true * sqaure_pred + (1 - y_true) * margin_square)\n",
    "\n",
    "\n",
    "def create_pairs(x, digit_indices):\n",
    "    '''Positive and negative pair creation.\n",
    "    Alternates between positive and negative pairs.\n",
    "    '''\n",
    "    pairs = []\n",
    "    labels = []\n",
    "    n = min([len(digit_indices[d]) for d in range(num_classes)]) - 1\n",
    "    for d in range(num_classes):\n",
    "        for i in range(n):\n",
    "            z1, z2 = digit_indices[d][i], digit_indices[d][i + 1]\n",
    "            pairs += [[x[z1], x[z2]]]\n",
    "            inc = random.randrange(1, num_classes)\n",
    "            dn = (d + inc) % num_classes\n",
    "            z1, z2 = digit_indices[d][i], digit_indices[dn][i]\n",
    "            pairs += [[x[z1], x[z2]]]\n",
    "            labels += [1, 0]\n",
    "    return np.array(pairs), np.array(labels)\n",
    "\n",
    "\n",
    "def create_base_network(input_shape):\n",
    "    '''Base network to be shared (eq. to feature extraction).\n",
    "    '''\n",
    "    input = Input(shape=input_shape)\n",
    "    x = Conv2D(1, (3, 3), strides = (1,1), padding = 'valid')(input)\n",
    "    x = Flatten()(x)\n",
    "    x = Dense(128, activation='relu')(x)\n",
    "    x = Dropout(0.1)(x)\n",
    "    x = Dense(128, activation='relu')(x)\n",
    "    x = Dropout(0.1)(x)\n",
    "    x = Dense(128, activation='relu')(x)\n",
    "    return Model(input, x)\n",
    "\n",
    "\n",
    "def compute_accuracy(y_true, y_pred):\n",
    "    '''Compute classification accuracy with a fixed threshold on distances.\n",
    "    '''\n",
    "    pred = y_pred.ravel() < 0.5\n",
    "    return np.mean(pred == y_true)\n",
    "\n",
    "\n",
    "def accuracy(y_true, y_pred):\n",
    "    '''Compute classification accuracy with a fixed threshold on distances.\n",
    "    '''\n",
    "    return K.mean(K.equal(y_true, K.cast(y_pred < 0.5, y_true.dtype)))\n",
    "\n",
    "\n",
    "# the data, split between train and test sets\n",
    "(x_train, y_train), (x_test, y_test) = mnist.load_data()\n",
    "x_train = x_train.astype('float32')\n",
    "x_test = x_test.astype('float32')\n",
    "x_train /= 255\n",
    "x_test /= 255\n",
    "input_shape = x_train.shape[1:] # (28, 28)\n",
    "\n",
    "# add chanel to input_shape, (28, 28) => (28, 28, 1) which satify to My model ResNet50\n",
    "# tuple operation\n",
    "chanel = (1,)\n",
    "input_shape = input_shape + chanel\n",
    "\n",
    "# create training+test positive and negative pairs\n",
    "digit_indices = [np.where(y_train == i)[0] for i in range(num_classes)]\n",
    "tr_pairs, tr_y = create_pairs(x_train, digit_indices)\n",
    "\n",
    "#expend dims, expend data dims form (samples, 2, 28, 28) => (samples, 2, 28, 28, 1)\n",
    "tr_pairs = np.expand_dims(tr_pairs, axis=4)\n",
    "\n",
    "#print(tr_pairs.shape)\n",
    "digit_indices = [np.where(y_test == i)[0] for i in range(num_classes)]\n",
    "te_pairs, te_y = create_pairs(x_test, digit_indices)\n",
    "\n",
    "#expend dims, expend data dims form (samples, 2, 28, 28) => (samples, 2, 28, 28, 1), same to tr_pairs\n",
    "te_pairs = np.expand_dims(te_pairs, axis=4)\n",
    "\n",
    "# network definition\n",
    "#base_network = create_base_network(input_shape)\n",
    "#base_network.summary()\n",
    "\n",
    "base_network = ResNet50(input_shape = input_shape, classes = 128)\n",
    "base_network.summary()\n",
    "input_a = Input(shape=input_shape) # Tensor(\"input_27:0\", shape=(?, 28, 28, 1), dtype=float32)\n",
    "input_b = Input(shape=input_shape)\n",
    "#print(input_a)\n",
    "\n",
    "processed_a = base_network(input_a)\n",
    "processed_b = base_network(input_b)\n",
    "\n",
    "distance = Lambda(euclidean_distance,\n",
    "                  output_shape=eucl_dist_output_shape)([processed_a, processed_b])\n",
    "\n",
    "model = Model([input_a, input_b], distance)\n",
    "\n",
    "# train\n",
    "rms = RMSprop()\n",
    "model.compile(loss=contrastive_loss, optimizer=rms, metrics=[accuracy])\n",
    "model.fit([tr_pairs[:, 0], tr_pairs[:, 1]], tr_y,\n",
    "          batch_size=128,\n",
    "          epochs=epochs,\n",
    "          validation_data=([te_pairs[:, 0], te_pairs[:, 1]], te_y))\n",
    "\n",
    "# compute final accuracy on training and test sets\n",
    "y_pred = model.predict([tr_pairs[:, 0], tr_pairs[:, 1]])\n",
    "tr_acc = compute_accuracy(tr_y, y_pred)\n",
    "y_pred = model.predict([te_pairs[:, 0], te_pairs[:, 1]])\n",
    "te_acc = compute_accuracy(te_y, y_pred)\n",
    "\n",
    "print('* Accuracy on training set: %0.2f%%' % (100 * tr_acc))\n",
    "print('* Accuracy on test set: %0.2f%%' % (100 * te_acc))"
   ]
  },
  {
   "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": "3.6.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}