Simple Convnet - MNIST

In [1]:

    

WEIGHTS_FILEPATH = 'mnist_cnn.hdf5'
MODEL_ARCH_FILEPATH = 'mnist_cnn.json'


    

In [2]:

    

import numpy as np
np.random.seed(1337)  # for reproducibility

from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten, Activation
from keras.layers import Conv2D, MaxPooling2D
from keras.utils import np_utils
from keras.callbacks import EarlyStopping, ModelCheckpoint


    

    




Using TensorFlow backend.

In [3]:

    

num_classes = 10

# input image dimensions
img_rows, img_cols = 28, 28

# the data, shuffled and split between train and test sets
(x_train, y_train), (x_test, y_test) = mnist.load_data()

x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 1)
x_test = x_test.reshape(x_test.shape[0], img_rows, img_cols, 1)
input_shape = (img_rows, img_cols, 1)

x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_train /= 255
x_test /= 255
print('x_train shape:', x_train.shape)
print(x_train.shape[0], 'train samples')
print(x_test.shape[0], 'test samples')

# convert class vectors to binary class matrices
y_train = np_utils.to_categorical(y_train, num_classes)
y_test = np_utils.to_categorical(y_test, num_classes)


    

    




x_train shape: (60000, 28, 28, 1)
60000 train samples
10000 test samples

In [4]:

    

# Sequential Model

model = Sequential()
model.add(Conv2D(32, (3, 3), input_shape=input_shape))
model.add(Activation('relu'))
model.add(Conv2D(32, (3, 3)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(128))
model.add(Activation('relu'))
model.add(Dropout(0.5))
model.add(Dense(num_classes))
model.add(Activation('softmax'))

model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])


    

In [5]:

    

# Model saving callback
checkpointer = ModelCheckpoint(filepath=WEIGHTS_FILEPATH, monitor='val_acc', verbose=1, save_best_only=True)

# Early stopping
early_stopping = EarlyStopping(monitor='val_acc', verbose=1, patience=5)

# Train
batch_size = 128
epochs = 100
model.fit(x_train, y_train, batch_size=batch_size, epochs=epochs, verbose=2,
          callbacks=[checkpointer, early_stopping], 
          validation_data=(x_test, y_test))
score = model.evaluate(x_test, y_test, verbose=0)
print('Test score:', score[0])
print('Test accuracy:', score[1])


    

    




Train on 60000 samples, validate on 10000 samples
Epoch 1/100
Epoch 00000: val_acc improved from -inf to 0.98000, saving model to mnist_cnn.hdf5
6s - loss: 0.2856 - acc: 0.9132 - val_loss: 0.0613 - val_acc: 0.9800
Epoch 2/100
Epoch 00001: val_acc improved from 0.98000 to 0.98770, saving model to mnist_cnn.hdf5
4s - loss: 0.1004 - acc: 0.9702 - val_loss: 0.0394 - val_acc: 0.9877
Epoch 3/100
Epoch 00002: val_acc improved from 0.98770 to 0.98850, saving model to mnist_cnn.hdf5
4s - loss: 0.0734 - acc: 0.9775 - val_loss: 0.0353 - val_acc: 0.9885
Epoch 4/100
Epoch 00003: val_acc improved from 0.98850 to 0.98940, saving model to mnist_cnn.hdf5
4s - loss: 0.0608 - acc: 0.9818 - val_loss: 0.0305 - val_acc: 0.9894
Epoch 5/100
Epoch 00004: val_acc improved from 0.98940 to 0.98990, saving model to mnist_cnn.hdf5
4s - loss: 0.0535 - acc: 0.9837 - val_loss: 0.0325 - val_acc: 0.9899
Epoch 6/100
Epoch 00005: val_acc improved from 0.98990 to 0.99170, saving model to mnist_cnn.hdf5
4s - loss: 0.0461 - acc: 0.9858 - val_loss: 0.0275 - val_acc: 0.9917
Epoch 7/100
Epoch 00006: val_acc did not improve
4s - loss: 0.0424 - acc: 0.9870 - val_loss: 0.0320 - val_acc: 0.9914
Epoch 8/100
Epoch 00007: val_acc did not improve
4s - loss: 0.0391 - acc: 0.9876 - val_loss: 0.0297 - val_acc: 0.9910
Epoch 9/100
Epoch 00008: val_acc did not improve
4s - loss: 0.0350 - acc: 0.9887 - val_loss: 0.0276 - val_acc: 0.9916
Epoch 10/100
Epoch 00009: val_acc did not improve
4s - loss: 0.0323 - acc: 0.9893 - val_loss: 0.0281 - val_acc: 0.9917
Epoch 11/100
Epoch 00010: val_acc did not improve
4s - loss: 0.0304 - acc: 0.9903 - val_loss: 0.0289 - val_acc: 0.9915
Epoch 12/100
Epoch 00011: val_acc improved from 0.99170 to 0.99200, saving model to mnist_cnn.hdf5
4s - loss: 0.0290 - acc: 0.9911 - val_loss: 0.0293 - val_acc: 0.9920
Epoch 13/100
Epoch 00012: val_acc improved from 0.99200 to 0.99290, saving model to mnist_cnn.hdf5
4s - loss: 0.0274 - acc: 0.9912 - val_loss: 0.0237 - val_acc: 0.9929
Epoch 14/100
Epoch 00013: val_acc did not improve
4s - loss: 0.0261 - acc: 0.9913 - val_loss: 0.0285 - val_acc: 0.9913
Epoch 15/100
Epoch 00014: val_acc did not improve
4s - loss: 0.0227 - acc: 0.9927 - val_loss: 0.0288 - val_acc: 0.9922
Epoch 16/100
Epoch 00015: val_acc did not improve
4s - loss: 0.0233 - acc: 0.9925 - val_loss: 0.0281 - val_acc: 0.9921
Epoch 17/100
Epoch 00016: val_acc did not improve
4s - loss: 0.0214 - acc: 0.9930 - val_loss: 0.0343 - val_acc: 0.9912
Epoch 18/100
Epoch 00017: val_acc did not improve
4s - loss: 0.0198 - acc: 0.9935 - val_loss: 0.0300 - val_acc: 0.9924
Epoch 19/100
Epoch 00018: val_acc did not improve
4s - loss: 0.0192 - acc: 0.9936 - val_loss: 0.0299 - val_acc: 0.9924
Epoch 00018: early stopping
Test score: 0.0298652117307
Test accuracy: 0.9924

In [6]:

    

with open(MODEL_ARCH_FILEPATH, 'w') as f:
    f.write(model.to_json())


    

In [ ]: