def create_model(session, sent_vocab_size, slot_vocab_size, intent_vocab_size):
"""Create model and initialize or load parameters in session."""
with tf.variable_scope("model_tylh", reuse=None):
model_train = multi_task_model.MultiTaskModel(
sent_vocab_size, slot_vocab_size, intent_vocab_size, FLAGS.max_sequence_length,
FLAGS.word_embedding_size, FLAGS.size, FLAGS.num_layers, FLAGS.max_gradient_norm, FLAGS.batch_size,
learning_rate=FLAGS.learning_rate, alpha=FLAGS.alpha,
dropout_keep_prob=FLAGS.dropout_keep_prob, use_lstm=True,
forward_only=False)
# use_attention=FLAGS.use_attention,
# bidirectional_rnn=FLAGS.bidirectional_rnn,
# task=task)
with tf.variable_scope("model_tylh", reuse=True):
model_test = multi_task_model.MultiTaskModel(
sent_vocab_size, slot_vocab_size, intent_vocab_size, FLAGS.max_sequence_length,
FLAGS.word_embedding_size, FLAGS.size, FLAGS.num_layers, FLAGS.max_gradient_norm, FLAGS.batch_size,
learning_rate=FLAGS.learning_rate, alpha=FLAGS.alpha,
dropout_keep_prob=FLAGS.dropout_keep_prob, use_lstm=True,
forward_only=True)
# use_attention=FLAGS.use_attention,
# bidirectional_rnn=FLAGS.bidirectional_rnn,
# task=task)
ckpt = tf.train.get_checkpoint_state(FLAGS.train_dir)
if ckpt and tf.gfile.Exists(ckpt.model_checkpoint_path):
print("Reading model parameters from %s" % ckpt.model_checkpoint_path)
model_train.saver.restore(session, ckpt.model_checkpoint_path)
else:
print("Created model with fresh parameters.")
session.run(tf.global_variables_initializer())
return model_train, model_test
def create_model(session, source_vocab_size, target_vocab_size, label_vocab_size):
"""Create model and initialize or load parameters in session."""
with tf.variable_scope("model", reuse=None):
model_train = multi_task_model.MultiTaskModel(
source_vocab_size, target_vocab_size, label_vocab_size, _buckets,
FLAGS.word_embedding_size, FLAGS.size, FLAGS.num_layers, FLAGS.max_gradient_norm, FLAGS.batch_size,
dropout_keep_prob=FLAGS.dropout_keep_prob, use_lstm=True,
forward_only=False,
use_attention=FLAGS.use_attention,
bidirectional_rnn=FLAGS.bidirectional_rnn,
task=task)
with tf.variable_scope("model", reuse=True):
model_test = multi_task_model.MultiTaskModel(
source_vocab_size, target_vocab_size, label_vocab_size, _buckets,
FLAGS.word_embedding_size, FLAGS.size, FLAGS.num_layers, FLAGS.max_gradient_norm, FLAGS.batch_size,
dropout_keep_prob=FLAGS.dropout_keep_prob, use_lstm=True,
forward_only=True,
use_attention=FLAGS.use_attention,
bidirectional_rnn=FLAGS.bidirectional_rnn,
task=task)
ckpt = tf.train.get_checkpoint_state(FLAGS.train_dir)
if ckpt and tf.gfile.Exists(ckpt.model_checkpoint_path):
print("Reading model parameters from %s" % ckpt.model_checkpoint_path)
model_train.saver.restore(session, ckpt.model_checkpoint_path)
else:
print("Created model with fresh parameters.")
session.run(tf.initialize_all_variables())
return model_train, model_test
def create_model(session, source_vocab_size, target_vocab_size,
label_vocab_size):
"""Create model and initialize or load parameters in session."""
with tf.variable_scope("model", reuse=None):
model_train = multi_task_model.MultiTaskModel(
source_vocab_size,
target_vocab_size,
label_vocab_size,
_buckets,
FLAGS.word_embedding_size,
FLAGS.size,
FLAGS.num_layers,
FLAGS.max_gradient_norm,
FLAGS.batch_size,
dropout_keep_prob=FLAGS.dropout_keep_prob,
use_lstm=True,
forward_only=False,
use_attention=FLAGS.use_attention,
bidirectional_rnn=FLAGS.bidirectional_rnn,
task=task)
with tf.variable_scope("model", reuse=True):
model_test = multi_task_model.MultiTaskModel(
source_vocab_size,
target_vocab_size,
label_vocab_size,
_buckets,
FLAGS.word_embedding_size,
FLAGS.size,
FLAGS.num_layers,
FLAGS.max_gradient_norm,
FLAGS.batch_size,
dropout_keep_prob=FLAGS.dropout_keep_prob,
use_lstm=True,
forward_only=True,
use_attention=FLAGS.use_attention,
bidirectional_rnn=FLAGS.bidirectional_rnn,
task=task)
ckpt = tf.train.get_checkpoint_state(FLAGS.train_dir)
if ckpt:
tf.logging.info("Reading model parameters from %s" %
ckpt.model_checkpoint_path)
# saver = tf.train.Saver(max_to_keep=3)
# model_file = tf.train.latest_checkpoint('model_tmp/')
# saver.restore(session, model_file)
init_op = tf.group(tf.global_variables_initializer(),
tf.local_variables_initializer())
session.run(init_op)
model_train.saver.restore(session, ckpt.model_checkpoint_path)
else:
tf.logging.info("Created model with fresh parameters.")
session.run(tf.global_variables_initializer())
return model_train, model_test
def create_model(session, source_vocab_size, target_vocab_size,
label_vocab_size, lm_vocab_size):
"""Create model and initialize or load parameters in session."""
with tf.variable_scope("model", reuse=None):
model_train = multi_task_model.MultiTaskModel(
source_vocab_size,
lm_vocab_size,
target_vocab_size,
label_vocab_size,
_buckets,
FLAGS.word_embedding_size,
FLAGS.size,
FLAGS.num_layers,
FLAGS.max_gradient_norm,
FLAGS.batch_size,
dropout_keep_prob=FLAGS.dropout_keep_prob,
use_lstm=True,
lm_cost_weight=FLAGS.lm_cost_weight,
forward_only=False,
DNN_at_output=FLAGS.DNN_at_output,
dnn_hidden_layer_size=FLAGS.dnn_hidden_layer_size,
output_emb_size=FLAGS.output_emb_size,
zero_intent_thres=FLAGS.zero_intent_thres)
with tf.variable_scope("model", reuse=True):
model_test = multi_task_model.MultiTaskModel(
source_vocab_size,
lm_vocab_size,
target_vocab_size,
label_vocab_size,
_buckets,
FLAGS.word_embedding_size,
FLAGS.size,
FLAGS.num_layers,
FLAGS.max_gradient_norm,
FLAGS.batch_size,
dropout_keep_prob=FLAGS.dropout_keep_prob,
use_lstm=True,
lm_cost_weight=FLAGS.lm_cost_weight,
forward_only=True,
DNN_at_output=FLAGS.DNN_at_output,
dnn_hidden_layer_size=FLAGS.dnn_hidden_layer_size,
output_emb_size=FLAGS.output_emb_size,
zero_intent_thres=FLAGS.zero_intent_thres)
ckpt = tf.train.get_checkpoint_state(FLAGS.train_dir)
if ckpt and tf.gfile.Exists(ckpt.model_checkpoint_path):
print("Reading model parameters from %s" % ckpt.model_checkpoint_path)
model_train.saver.restore(session, ckpt.model_checkpoint_path)
else:
print("Created model with fresh parameters.")
session.run(tf.global_variables_initializer())
return model_train, model_test
def create_model(session, source_vocab_size, target_vocab_size,
label_vocab_size):
# 4-2-1. Create train model.
with tf.variable_scope("model", reuse=None):
model_train = multi_task_model.MultiTaskModel(
source_vocab_size,
target_vocab_size,
label_vocab_size,
_buckets,
FLAGS.word_embedding_size,
FLAGS.size,
FLAGS.num_layers,
FLAGS.max_gradient_norm,
FLAGS.batch_size,
dropout_keep_prob=FLAGS.dropout_keep_prob,
use_lstm=True,
forward_only=False,
use_attention=FLAGS.use_attention,
bidirectional_rnn=FLAGS.bidirectional_rnn,
task=task)
# 4-2-2. Create test model.
with tf.variable_scope("model", reuse=True):
model_test = multi_task_model.MultiTaskModel(
source_vocab_size,
target_vocab_size,
label_vocab_size,
_buckets,
FLAGS.word_embedding_size,
FLAGS.size,
FLAGS.num_layers,
FLAGS.max_gradient_norm,
FLAGS.batch_size,
dropout_keep_prob=FLAGS.dropout_keep_prob,
use_lstm=True,
forward_only=True,
use_attention=FLAGS.use_attention,
bidirectional_rnn=FLAGS.bidirectional_rnn,
task=task)
# 4-2-3. Get "model paramters" or Initialize the model parameters.
# 4-2-3-1. Get model parameters.
ckpt = tf.train.get_checkpoint_state(FLAGS.train_dir)
if ckpt:
print("Reading model parameters from %s" % ckpt.model_checkpoint_path)
model_train.saver.restore(session, ckpt.model_checkpoint_path)
# 4-2-3-2. Initialize the model parameters.
else:
print("Created model with fresh parameters.")
session.run(tf.global_variables_initializer())
return model_train, model_test
def create_model(session, source_vocab_size, target_vocab_size,
label_vocab_size):
"""Create model and initialize or load parameters in session."""
with tf.variable_scope("model", reuse=None):
model_train = multi_task_model.MultiTaskModel(
source_vocab_size,
target_vocab_size,
label_vocab_size,
_buckets,
FLAGS.word_embedding_size,
FLAGS.size,
FLAGS.num_layers,
FLAGS.max_gradient_norm,
FLAGS.batch_size,
dropout_keep_prob=FLAGS.dropout_keep_prob,
use_lstm=True,
forward_only=False,
use_attention=FLAGS.use_attention,
bidirectional_rnn=FLAGS.bidirectional_rnn,
task=task)
with tf.variable_scope("model", reuse=True):
global model_test
model_test = multi_task_model.MultiTaskModel(
source_vocab_size,
target_vocab_size,
label_vocab_size,
_buckets,
FLAGS.word_embedding_size,
FLAGS.size,
FLAGS.num_layers,
FLAGS.max_gradient_norm,
FLAGS.batch_size,
dropout_keep_prob=FLAGS.dropout_keep_prob,
use_lstm=True,
forward_only=True,
use_attention=FLAGS.use_attention,
bidirectional_rnn=FLAGS.bidirectional_rnn,
task=task)
restorationPath = "./model_tmp/model_final.ckpt" # It will change. Somehow we must solve this problem
if True:
print("Reading model parameters from %s" % restorationPath)
model_train.saver.restore(session, restorationPath)
#model_test.saver.restore(session, restorationPath)
else:
print("Created model with fresh parameters.")
session.run(tf.initialize_all_variables())
return model_train, model_test
def create_model(session,
source_vocab_size,
target_vocab_size,
label_vocab_size):
"""创建模型并在会话中初始化或者加载参数。"""
with tf.variable_scope("model", reuse=None):
model_train = multi_task_model.MultiTaskModel(
source_vocab_size,
target_vocab_size,
label_vocab_size,
_buckets,
FLAGS.word_embedding_size,
FLAGS.size, FLAGS.num_layers,
FLAGS.max_gradient_norm,
FLAGS.batch_size,
dropout_keep_prob=FLAGS.dropout_keep_prob,
use_lstm=True,
forward_only=False,
use_attention=FLAGS.use_attention,
bidirectional_rnn=FLAGS.bidirectional_rnn,
task=task)
with tf.variable_scope("model", reuse=True):
model_test = multi_task_model.MultiTaskModel(
source_vocab_size,
target_vocab_size,
label_vocab_size,
_buckets,
FLAGS.word_embedding_size,
FLAGS.size,
FLAGS.num_layers,
FLAGS.max_gradient_norm,
FLAGS.batch_size,
dropout_keep_prob=FLAGS.dropout_keep_prob,
use_lstm=True,
forward_only=True,
use_attention=FLAGS.use_attention,
bidirectional_rnn=FLAGS.bidirectional_rnn,
task=task)
# 判断是否添加模型
checkpoint = tf.train.get_checkpoint_state(FLAGS.train_dir)
if checkpoint:
print("读取模型参数： %s" % checkpoint.model_checkpoint_path)
model_train.saver.restore(session, checkpoint.model_checkpoint_path)
else:
print("使用新的参数创建模型。")
session.run(tf.global_variables_initializer())
return model_train, model_test
def load_nlu_model():
session = tf.Session()
session.run(tf.global_variables_initializer())
with tf.variable_scope("model", reuse=None):
model_test = multi_task_model.MultiTaskModel(
sent_vocab_size,
slot_vocab_size,
intent_vocab_size,
FLAGS.max_sequence_length,
FLAGS.word_embedding_size,
FLAGS.size,
FLAGS.num_layers,
FLAGS.max_gradient_norm,
FLAGS.batch_size,
learning_rate=FLAGS.learning_rate,
alpha=FLAGS.alpha,
dropout_keep_prob=FLAGS.dropout_keep_prob,
use_lstm=True,
forward_only=True)
model_test.saver.restore(
session,
tf.train.get_checkpoint_state(FLAGS.train_dir).model_checkpoint_path)
return session, model_test
