# input image dimensions
img_rows, img_cols = 28, 28
chnls = 1
(x_train, y_train), (x_test, y_test) = dataset.load_data()
x_train = x_train.reshape(x_train.shape[0], chnls, img_rows, img_cols)
x_test = x_test.reshape(x_test.shape[0], chnls, img_rows, img_cols)
y_train = to_categorical(y_train, nb_classes)
y_test = to_categorical(y_test, nb_classes)
x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
y_train = y_train.astype('float32')
y_test = y_test.astype('float32')
x_train /= 255
x_test /= 255
print("Training...")
for epoch in range(nb_epoch):
print("Epoch: {}".format(epoch))
print("Training loss: {}".format(
train_epoch(x_train, y_train, train_func, batch_size)))
x_train, y_train = shuffle(x_train, y_train)
val_err, val_loss = test_epoch(x_test, y_test, test_func, batch_size)
print("Validation accuracy: {}".format((1 - val_err) * 100))
score = test_func(x_test, y_test)
params = lasagne.layers.get_all_param_values(model)
save_parameters(params, '{:2.2f}.h5'.format((1 - score[1]) * 100))
train_set.y = np.float32(np.eye(10)[train_set.y])
valid_set.y = np.float32(np.eye(10)[valid_set.y])
test_set.y = np.float32(np.eye(10)[test_set.y])
# for hinge loss
train_set.y = 2 * train_set.y - 1.
valid_set.y = 2 * valid_set.y - 1.
test_set.y = 2 * test_set.y - 1.
print('Building the CNN...')
cnn, train_fn, val_fn = build_network()
print('Training...')
binary_net.train(train_fn, val_fn, cnn, batch_size, LR_start, LR_decay,
num_epochs, train_set.X, train_set.y, valid_set.X,
valid_set.y, test_set.X, test_set.y,
shuffle_parts=shuffle_parts)
W = lasagne.layers.get_all_layers(cnn)[1].W.get_value()
import matplotlib.pyplot as plt
plt.hist(W.flatten())
plt.title("Weight distribution of first hidden convolution layer")
# Dump the network weights to a file
filepath = 'binarynet.h5'
params = lasagne.layers.get_all_param_values(cnn)
save_parameters(params, filepath)
train_set.y = np.float32(np.eye(10)[train_set.y])
valid_set.y = np.float32(np.eye(10)[valid_set.y])
test_set.y = np.float32(np.eye(10)[test_set.y])
# for hinge loss
train_set.y = 2 * train_set.y - 1.
valid_set.y = 2 * valid_set.y - 1.
test_set.y = 2 * test_set.y - 1.
print('Building the CNN...')
model, train_func, val_func = build_network()
print('Training...')
binary_connect.train(train_func, val_func, batch_size, LR_start, LR_decay,
num_epochs, train_set.X, train_set.y, valid_set.X,
valid_set.y, test_set.X, test_set.y)
W = lasagne.layers.get_all_layers(model)[1].W.get_value()
plt.hist(W.flatten())
plt.title("Weight distribution of first hidden convolution layer")
# Dump the network weights to a file
filepath = 'binaryconnect'
parameters = lasagne.layers.get_all_param_values(model)
save_parameters(parameters, filepath + '.h5')
params_b = lasagne.layers.get_all_params(model, binary=True)
save_parameters(parameters, filepath + '_binary.h5')