def mnist_lenet(input_shape, classes):
model = LeNet(input_shape, classes)
optimizer = keras.optimizers.SGD(lr=0.01)
model.compile(optimizer=optimizer,
loss=keras.losses.categorical_crossentropy,
metrics=['accuracy'])
return model
if __name__ == '__main__':
(x_train, y_train), (x_test, y_test) = load_mnist_data_metrics()
initial_epoch = 0
epochs = 20
batch_size = 2000
input_shape = (28, 28, 1)
num_classes = 10
import glob
ckpt_filelist = glob.glob('./ckpt_lenet/*.h5')
if len(ckpt_filelist) == 0:
model = LeNet(input_shape, num_classes)
model.compile(loss=losses.categorical_crossentropy,
optimizer=optimizers.Adadelta(),
metrics=['accuracy'])
else:
filename = ckpt_filelist[-1]
model = load_model(filename)
initial_epoch = int(filename.split('weights.')[-1].split('-')[0])
print('restore from ckpt, initial_echo = {0}'.format(initial_epoch))
print(model.summary())
tensorboard = TensorBoard(log_dir='./logs_lenet')
# weights.{epoch:02d}-{val_loss:.2f}.hdf5 : a kind of code rule
checkpointer = ModelCheckpoint(
filepath='./ckpt_lenet/weights.{epoch:02d}-{val_loss:.2f}.h5',
monitor='val_loss')
