def posterior_test_multitask(session,
posteriors_op_main,
posteriors_op_second,
network,
dataset,
label_type_second,
save_path=None,
show=False):
"""Visualize label posteriors of the multi-task CTC model.
Args:
session: session of training model
posteriois_op_main: operation for computing posteriors in the main task
posteriois_op_second: operation for computing posteriors in the second
task
network: network to evaluate
dataset: An instance of a `Dataset` class
label_type_second: string, phone39 or phone48 or phone61
save_path: path to save ctc outpus
show: if True, show each figure
"""
# Batch size is expected to be 1
iteration = dataset.data_num
# Make data generator
mini_batch = dataset.next_batch(batch_size=1)
save_path = mkdir_join(save_path, 'ctc_output')
for step in range(iteration):
# Create feed dictionary for next mini batch
inputs, _, _, inputs_seq_len, input_names = mini_batch.__next__()
feed_dict = {
network.inputs: inputs,
network.inputs_seq_len: inputs_seq_len,
network.keep_prob_input: 1.0,
network.keep_prob_hidden: 1.0
}
# Visualize
max_frame_num = inputs.shape[1]
posteriors_char = session.run(posteriors_op_main, feed_dict=feed_dict)
posteriors_phone = session.run(posteriors_op_second,
feed_dict=feed_dict)
i_batch = 0 # index in mini-batch
posteriors_index = np.array(
[i_batch * max_frame_num + i for i in range(max_frame_num)])
plot_probs_ctc_char_phone(
probs_char=posteriors_char[posteriors_index]
[:int(inputs_seq_len[0]), :],
probs_phone=posteriors_phone[posteriors_index]
[:int(inputs_seq_len[0]), :],
wav_index=input_names[0],
label_type_second=label_type_second,
save_path=save_path,
show=show)
def posterior_test(session, posteriors_op, network, dataset, label_type,
save_path=None, show=False):
"""Visualize label posteriors of CTC model.
Args:
session: session of training model
posteriois_op: operation for computing posteriors
network: network to evaluate
dataset: An instance of a `Dataset` class
label_type: string, kanji or kana or phone
save_path: path to save ctc outputs
show: if True, show each figure
"""
# Batch size is expected to be 1
iteration = dataset.data_num
# Make data generator
mini_batch = dataset.next_batch(batch_size=1)
save_path = mkdir_join(save_path, 'ctc_output')
for step in range(iteration):
# Create feed dictionary for next mini batch
inputs, _, inputs_seq_len, input_names = mini_batch.__next__()
feed_dict = {
network.inputs: inputs,
network.inputs_seq_len: inputs_seq_len,
network.keep_prob_input: 1.0,
network.keep_prob_hidden: 1.0,
network.keep_prob_output: 1.0
}
# Visualize
max_frame_num = inputs.shape[1]
posteriors = session.run(posteriors_op, feed_dict=feed_dict)
i_batch = 0 # index in mini-batch
posteriors_index = np.array(
[i_batch * max_frame_num + i for i in range(max_frame_num)])
plot_probs_ctc(
probs=posteriors[posteriors_index][:int(inputs_seq_len[0]), :],
wav_index=input_names[0],
label_type=label_type,
save_path=save_path,
show=show)
def main(config_path):
# Load a config file (.yml)
with open(config_path, "r") as f:
config = yaml.load(f)
param = config['param']
# Except for a blank label
if param['label_type'] == 'kanji':
param['num_classes'] = 3386
elif param['label_type'] == 'kana':
param['num_classes'] = 147
elif param['label_type'] == 'phone':
param['num_classes'] = 38
# Model setting
CTCModel = load(model_type=param['model'])
network = CTCModel(batch_size=param['batch_size'],
input_size=param['input_size'] * param['num_stack'],
num_unit=param['num_unit'],
num_layer=param['num_layer'],
bottleneck_dim=param['bottleneck_dim'],
num_classes=param['num_classes'],
parameter_init=param['weight_init'],
clip_grad=param['clip_grad'],
clip_activation=param['clip_activation'],
dropout_ratio_input=param['dropout_input'],
dropout_ratio_hidden=param['dropout_hidden'],
num_proj=param['num_proj'],
weight_decay=param['weight_decay'])
network.model_name = param['model']
network.model_name += '_' + str(param['num_unit'])
network.model_name += '_' + str(param['num_layer'])
network.model_name += '_' + param['optimizer']
network.model_name += '_lr' + str(param['learning_rate'])
if param['bottleneck_dim'] != 0:
network.model_name += '_bottoleneck' + str(param['bottleneck_dim'])
if param['num_proj'] != 0:
network.model_name += '_proj' + str(param['num_proj'])
if param['num_stack'] != 1:
network.model_name += '_stack' + str(param['num_stack'])
if param['weight_decay'] != 0:
network.model_name += '_weightdecay' + str(param['weight_decay'])
if param['train_data_size'] == 'large':
network.model_name += '_large'
# Set save path
network.model_dir = mkdir('/n/sd8/inaguma/result/csj/')
network.model_dir = mkdir_join(network.model_dir, 'ctc')
network.model_dir = mkdir_join(network.model_dir, param['label_type'])
network.model_dir = mkdir_join(network.model_dir, network.model_name)
# Reset model directory
if not isfile(join(network.model_dir, 'complete.txt')):
tf.gfile.DeleteRecursively(network.model_dir)
tf.gfile.MakeDirs(network.model_dir)
else:
raise ValueError('File exists.')
# Set process name
setproctitle('csj_ctc_' + param['label_type'] + '_' +
param['train_data_size'])
# Save config file
shutil.copyfile(config_path, join(network.model_dir, 'config.yml'))
sys.stdout = open(join(network.model_dir, 'train.log'), 'w')
print(network.model_name)
do_train(network=network, param=param)
sys.stdout = sys.__stdout__
def main(config_path):
# Load a config file (.yml)
with open(config_path, "r") as f:
config = yaml.load(f)
param = config['param']
if param['label_type_sub'] == 'phone61':
param['num_classes_sub'] = 61
elif param['label_type_sub'] == 'phone48':
param['num_classes_sub'] = 48
elif param['label_type_sub'] == 'phone39':
param['num_classes_sub'] = 39
# Model setting
CTCModel = load(model_type=param['model'])
network = CTCModel(batch_size=param['batch_size'],
input_size=param['input_size'] * param['num_stack'],
num_unit=param['num_unit'],
num_layer_main=param['num_layer_main'],
num_layer_sub=param['num_layer_sub'],
num_classes_main=33,
num_classes_sub=param['num_classes_sub'],
main_task_weight=param['main_task_weight'],
parameter_init=param['weight_init'],
clip_grad=param['clip_grad'],
clip_activation=param['clip_activation'],
dropout_ratio_input=param['dropout_input'],
dropout_ratio_hidden=param['dropout_hidden'],
num_proj=param['num_proj'],
weight_decay=param['weight_decay'])
network.model_name = param['model']
network.model_name += '_' + str(param['num_unit'])
network.model_name += '_main' + str(param['num_layer_main'])
network.model_name += '_sub' + str(param['num_layer_sub'])
network.model_name += '_' + param['optimizer']
network.model_name += '_lr' + str(param['learning_rate'])
if param['num_proj'] != 0:
network.model_name += '_proj' + str(param['num_proj'])
if param['dropout_input'] != 1:
network.model_name += '_dropi' + str(param['dropout_input'])
if param['dropout_hidden'] != 1:
network.model_name += '_droph' + str(param['dropout_hidden'])
if param['num_stack'] != 1:
network.model_name += '_stack' + str(param['num_stack'])
if param['weight_decay'] != 0:
network.model_name += '_weightdecay' + str(param['weight_decay'])
network.model_name += '_taskweight' + str(param['main_task_weight'])
if param['decay_rate'] != 1:
network.model_name += '_lrdecay' + \
str(param['decay_steps'] + param['decay_rate'])
# Set save path
network.model_dir = mkdir('/n/sd8/inaguma/result/timit/')
network.model_dir = mkdir_join(network.model_dir, 'ctc')
network.model_dir = mkdir_join(network.model_dir,
'char_' + param['label_type_sub'])
network.model_dir = mkdir_join(network.model_dir, network.model_name)
# Reset model directory
if not isfile(join(network.model_dir, 'complete.txt')):
tf.gfile.DeleteRecursively(network.model_dir)
tf.gfile.MakeDirs(network.model_dir)
else:
raise ValueError('File exists.')
# Set process name
setproctitle('timit_multictc')
# Save config file
shutil.copyfile(config_path, join(network.model_dir, 'config.yml'))
sys.stdout = open(join(network.model_dir, 'train.log'), 'w')
print(network.model_name)
do_train(network=network, param=param)
def main(config_path):
# Load a config file (.yml)
with open(config_path, "r") as f:
config = yaml.load(f)
param = config['param']
if param['label_type'] == 'phone61':
param['att_num_classes'] = 63
param['ctc_num_classes'] = 61
param['sos_index'] = 0
param['eos_index'] = 1
elif param['label_type'] == 'phone48':
param['att_num_classes'] = 50
param['ctc_num_classes'] = 48
param['sos_index'] = 0
param['eos_index'] = 1
elif param['label_type'] == 'phone39':
param['att_num_classes'] = 41
param['ctc_num_classes'] = 39
param['sos_index'] = 0
param['eos_index'] = 1
elif param['label_type'] == 'character':
param['att_num_classes'] = 35
param['ctc_num_classes'] = 33
param['sos_index'] = 1
param['eos_index'] = 2
# Model setting
# AttentionModel = load(model_type=config['model_name'])
network = JointCTCAttention(
batch_size=param['batch_size'],
input_size=param['input_size'],
encoder_num_unit=param['encoder_num_unit'],
encoder_num_layer=param['encoder_num_layer'],
attention_dim=param['attention_dim'],
attention_type=param['attention_type'],
decoder_num_unit=param['decoder_num_unit'],
decoder_num_layer=param['decoder_num_layer'],
embedding_dim=param['embedding_dim'],
att_num_classes=param['att_num_classes'],
ctc_num_classes=param['ctc_num_classes'],
att_task_weight=param['att_task_weight'],
sos_index=param['sos_index'],
eos_index=param['eos_index'],
max_decode_length=param['max_decode_length'],
# attention_smoothing=param['attention_smoothing'],
attention_weights_tempareture=param['attention_weights_tempareture'],
logits_tempareture=param['logits_tempareture'],
parameter_init=param['weight_init'],
clip_grad=param['clip_grad'],
clip_activation_encoder=param['clip_activation_encoder'],
clip_activation_decoder=param['clip_activation_decoder'],
dropout_ratio_input=param['dropout_input'],
dropout_ratio_hidden=param['dropout_hidden'],
weight_decay=param['weight_decay'])
network.model_name = param['model']
network.model_name += '_encoder' + str(param['encoder_num_unit'])
network.model_name += '_' + str(param['encoder_num_layer'])
network.model_name += '_attdim' + str(param['attention_dim'])
network.model_name += '_decoder' + str(param['decoder_num_unit'])
network.model_name += '_' + str(param['decoder_num_layer'])
network.model_name += '_' + param['optimizer']
network.model_name += '_lr' + str(param['learning_rate'])
network.model_name += '_' + param['attention_type']
# if bool(param['attention_smoothing']):
# network.model_name += '_smoothing'
if param['attention_weights_tempareture'] != 1:
network.model_name += '_sharpening' + \
str(param['attention_weights_tempareture'])
if param['weight_decay'] != 0:
network.model_name += '_weightdecay' + str(param['weight_decay'])
# Set save path
network.model_dir = mkdir('/n/sd8/inaguma/result/timit/')
network.model_dir = mkdir_join(network.model_dir, 'attention')
network.model_dir = mkdir_join(network.model_dir, param['label_type'])
network.model_dir = mkdir_join(network.model_dir, network.model_name)
# Reset model directory
if not isfile(join(network.model_dir, 'complete.txt')):
tf.gfile.DeleteRecursively(network.model_dir)
tf.gfile.MakeDirs(network.model_dir)
else:
raise ValueError('File exists.')
# Set process name
setproctitle('timit_jointctcatt_' + param['label_type'])
# Save config file
shutil.copyfile(config_path, join(network.model_dir, 'config.yml'))
sys.stdout = open(join(network.model_dir, 'train.log'), 'w')
print(network.model_name)
do_train(network=network, param=param)