save_arch(model, arch_path) # モデルを保存しておく
# トレーニングの準備
checkpoint_collback = ModelCheckpoint(filepath=weights_path,
monitor='val_loss',
save_best_only=True,
mode='auto')
change_lr = LearningRateScheduler(lambda epoch: float(learning_rates[epoch]))
flip_gen = FlippedImageDataGenerator()
sgd = SGD(lr=lr, momentum=momentum, nesterov=nesterov)
model.compile(loss=loss_method, optimizer=sgd)
# トレーニング
start_time = time.time()
print('start_time: %s' % (datetime.now()))
hist = model.fit_generator(flip_gen.flow(X_train, y_train),
samples_per_epoch=X_train.shape[0],
nb_epoch=nb_epoch,
validation_data=(X_val, y_val),
callbacks=[checkpoint_collback, change_lr])
print('end_time: %s, duracion(min): %d' %
(datetime.now(), int(time.time() - start_time) / 60))
# プロットしてファイルとして保存する
# plot_hist(hist, model_name)
# plot_model_arch(model, model_name)
save_history(hist, model_name)
save_arch(model, arch_path) # モデルを保存しておく
# トレーニングの準備
sgd = SGD(lr=start, momentum=momentum, nesterov=nesterov)
model.compile(loss=loss_method, optimizer=sgd)
#plot(model, to_file="model_{}.png".format(cols[0]), show_shapes=True)
flipgen = FlippedImageDataGenerator()
flipgen.flip_indices = setting['flip_indices']
early_stop = EarlyStopping(patience=patience)
learning_rates = np.linspace(start, stop, nb_epoch)
change_lr = LearningRateScheduler(
lambda epoch: float(learning_rates[epoch]))
weights_path = 'model/' + model_name + '-' + setting[
'id'] + '-weights-' + str(nb_epoch) + '.hdf5'
checkpoint_collback = ModelCheckpoint(filepath=weights_path,
monitor='val_loss',
save_best_only=True,
mode='auto')
print("Training model for columns {} for {} epochs".format(cols, nb_epoch))
# トレーニング実施
hist = model.fit_generator(
flipgen.flow(X_train, y_train),
samples_per_epoch=X_train.shape[0],
nb_epoch=nb_epoch,
validation_data=(X_val, y_val),
callbacks=[checkpoint_collback, change_lr, early_stop])
save_history(hist, model_name + '-' + setting['id'])