def take_action(self, parsed_args):
encoder_architecture = RNNArchitecture(
num_layers=parsed_args.num_layers,
num_units=parsed_args.num_units,
bidirectional=parsed_args.bidirectional_encoder,
cell_type=parsed_args.cell)
decoder_architecture = RNNArchitecture(
num_layers=parsed_args.num_layers,
num_units=parsed_args.num_units,
bidirectional=parsed_args.bidirectional_decoder,
cell_type=parsed_args.cell)
wrapper = FrequencyAutoencoderWrapper()
if not parsed_args.continue_training:
wrapper.initialize_model(
feature_shape=self.feature_shape,
model_filename=self.model_filename,
encoder_architecture=encoder_architecture,
decoder_architecture=decoder_architecture,
frequency_window_width=parsed_args.freq_window_width,
frequency_window_overlap=parsed_args.freq_window_overlap)
wrapper.train_model(
model_filename=self.model_filename,
record_files=self.record_files,
feature_shape=self.feature_shape,
num_instances=self.num_instances,
num_epochs=parsed_args.num_epochs,
batch_size=parsed_args.batch_size,
checkpoints_to_keep=parsed_args.checkpoints_to_keep,
learning_rate=parsed_args.learning_rate,
keep_prob=parsed_args.keep_prob,
encoder_noise=parsed_args.encoder_noise,
decoder_feed_previous_prob=parsed_args.feed_previous_prob)
def take_action(self, parsed_args):
encoder_architecture = RNNArchitecture(
num_layers=parsed_args.num_layers,
num_units=parsed_args.num_units,
bidirectional=parsed_args.bidirectional_encoder,
cell_type=parsed_args.cell)
decoder_architecture = RNNArchitecture(
num_layers=parsed_args.num_layers,
num_units=parsed_args.num_units,
bidirectional=parsed_args.bidirectional_decoder,
cell_type=parsed_args.cell)
wrapper = TimeAutoencoderWrapper()
if not parsed_args.continue_training:
wrapper.initialize_model(feature_shape=self.feature_shape,
model_filename=self.model_filename,
encoder_architecture=encoder_architecture,
decoder_architecture=decoder_architecture,
mask_silence=parsed_args.mask_silence)
wrapper.train_model(
model_filename=self.model_filename,
record_files=self.record_files,
feature_shape=self.feature_shape,
num_instances=self.num_instances,
num_epochs=parsed_args.num_epochs,
batch_size=parsed_args.batch_size,
learning_rate=parsed_args.learning_rate,
keep_prob=parsed_args.keep_prob,
encoder_noise=parsed_args.encoder_noise,
decoder_feed_previous_prob=parsed_args.feed_previous_prob)