def predict(self, test_x, out_scaler, gen_test_file_list):
#### compute predictions ####
io_funcs = BinaryIOCollection()
test_id_list = test_x.keys()
test_id_list.sort()
inference_batch_size = len(test_id_list)
test_file_number = len(test_id_list)
with tf.Session(graph=self.graph) as sess:
new_saver = tf.train.import_meta_graph(self.ckpt_dir,
"mymodel.ckpt.meta")
"""Notice change targets=tf.get_collection("targets")[0]"""
inputs_data = self.graph.get_collection("inputs_data")[0]
"""Notice Change decoder_outputs=tf.get_collection("decoder_outputs")[0]"""
inputs_sequence_length = self.graph.get_collection(
"inputs_sequence_length")[0]
target_sequence_length = self.graph.get_collection(
"target_sequence_length")[0]
print("loading the model parameters...")
new_saver.restore(sess, os.path.join(self.ckpt_dir,
"mymodel.ckpt"))
print("Model parameters are successfully restored")
print("generating features on held-out test data...")
for utt_index in range(test_file_number):
gen_test_file_name = gen_test_file_list[utt_index]
temp_test_x = test_x[test_id_list[utt_index]]
num_of_rows = temp_test_x.shape[0]
#utt_length=[len(utt) for utt in test_x.values()]
#max_step=max(utt_length)
temp_test_x = tf.reshape(temp_test_x,
[1, num_of_rows, self.n_in])
outputs = np.zeros(shape=[len(test_x), max_step, self.n_out],
dtype=np.float32)
#dec_cell=self.graph.get_collection("decoder_cell")[0]
print("Generating speech parameters ...")
for t in range(num_of_rows):
# outputs=sess.run(inference_output,{inputs_data:temp_test_x,inputs_sequence_length:utt_length,\
# target_sequence_length:utt_length})
_outputs=sess.run(decoder_outputs,feed_dict={inputs_data:temp_test_x,targets:outputs,inputs_sequence_length:[num_of_rows],\
target_sequence_length:[num_of_rows]})
# #print _outputs[:,t,:]
outputs[:, t, :] = _outputs[:, t, :]
data_utils.denorm_data(outputs, out_scaler)
io_funcs.array_to_binary_file(outputs, gen_test_file_name)
data_utils.drawProgressBar(utt_index + 1, test_file_number)
def predict(self, test_x, out_scaler, gen_test_file_list, sequential_training=False, stateful=False):
#### compute predictions ####
io_funcs = BinaryIOCollection()
test_id_list = test_x.keys()
test_id_list.sort()
test_file_number = len(test_id_list)
print("generating features on held-out test data...")
with tf.Session() as sess:
new_saver=tf.train.import_meta_graph(os.path.join(self.ckpt_dir,"mymodel.ckpt.meta"))
print "loading the model parameters..."
output_layer=tf.get_collection("output_layer")[0]
input_layer=tf.get_collection("input_layer")[0]
new_saver.restore(sess,os.path.join(self.ckpt_dir,"mymodel.ckpt"))
print "The model parameters are successfully restored"
for utt_index in xrange(test_file_number):
gen_test_file_name = gen_test_file_list[utt_index]
temp_test_x = test_x[test_id_list[utt_index]]
num_of_rows = temp_test_x.shape[0]
if not sequential_training:
is_training_batch=tf.get_collection("is_training_batch")[0]
if self.dropout_rate!=0.0:
is_training_drop=tf.get_collection("is_training_drop")[0]
y_predict=sess.run(output_layer,feed_dict={input_layer:temp_test_x,is_training_drop:False,is_training_batch:False})
else:
y_predict=sess.run(output_layer,feed_dict={input_layer:temp_test_x,is_training_batch:False})
else:
temp_test_x=np.reshape(temp_test_x,[1,num_of_rows,self.n_in])
hybrid=0
utt_length_placeholder=tf.get_collection("utt_length")[0]
if "tanh" in self.hidden_layer_type:
hybrid=1
is_training_batch=tf.get_collection("is_training_batch")[0]
if self.dropout_rate!=0.0:
is_training_drop=tf.get_collection("is_training_drop")[0]
if hybrid:
y_predict=sess.run(output_layer,feed_dict={input_layer:temp_test_x,utt_length_placeholder:[num_of_rows],is_training_drop:False,is_training_batch:False})
else:
y_predict=sess.run(output_layer,feed_dict={input_layer:temp_test_x,utt_length_placeholder:[num_of_rows],is_training_drop:False})
elif hybrid:
y_predict=sess.run(output_layer,feed_dict={input_layer:temp_test_x,utt_length_placeholder:[num_of_rows],is_training_batch:False})
else:
y_predict=sess.run(output_layer,feed_dict={input_layer:temp_test_x,utt_length_placeholder:[num_of_rows]})
data_utils.denorm_data(y_predict, out_scaler)
io_funcs.array_to_binary_file(y_predict, gen_test_file_name)
data_utils.drawProgressBar(utt_index+1, test_file_number)
def predict(self,test_x, out_scaler, gen_test_file_list):
#### compute predictions ####
io_funcs = BinaryIOCollection()
test_id_list = test_x.keys()
test_id_list.sort()
inference_batch_size=len(test_id_list)
test_file_number = len(test_id_list)
with tf.Session(graph=self.graph) as sess:
new_saver=tf.train.import_meta_graph(self.ckpt_dir,"mymodel.ckpt.meta")
"""Notice change targets=tf.get_collection("targets")[0]"""
inputs_data=self.graph.get_collection("inputs_data")[0]
"""Notice Change decoder_outputs=tf.get_collection("decoder_outputs")[0]"""
inputs_sequence_length=self.graph.get_collection("inputs_sequence_length")[0]
target_sequence_length=self.graph.get_collection("target_sequence_length")[0]
print "loading the model parameters..."
new_saver.restore(sess,os.path.join(self.ckpt_dir,"mymodel.ckpt"))
print "Model parameters are successfully restored"
print("generating features on held-out test data...")
for utt_index in xrange(test_file_number):
gen_test_file_name = gen_test_file_list[utt_index]
temp_test_x = test_x[test_id_list[utt_index]]
num_of_rows = temp_test_x.shape[0]
#utt_length=[len(utt) for utt in test_x.values()]
#max_step=max(utt_length)
temp_test_x = tf.reshape(temp_test_x,[1,num_of_rows,self.n_in])
outputs=np.zeros(shape=[len(test_x),max_step,self.n_out],dtype=np.float32)
#dec_cell=self.graph.get_collection("decoder_cell")[0]
print "Generating speech parameters ..."
for t in range(num_of_rows):
# outputs=sess.run(inference_output,{inputs_data:temp_test_x,inputs_sequence_length:utt_length,\
# target_sequence_length:utt_length})
_outputs=sess.run(decoder_outputs,feed_dict={inputs_data:temp_test_x,targets:outputs,inputs_sequence_length:[num_of_rows],\
target_sequence_length:[num_of_rows]})
# #print _outputs[:,t,:]
outputs[:,t,:]=_outputs[:,t,:]
data_utils.denorm_data(outputs, out_scaler)
io_funcs.array_to_binary_file(outputs, gen_test_file_name)
data_utils.drawProgressBar(utt_index+1, test_file_number)
def predict(self,
test_x,
out_scaler,
gen_test_file_list,
sequential_training=False,
stateful=False):
#### compute predictions ####
io_funcs = BinaryIOCollection()
test_id_list = list(test_x.keys())
test_id_list.sort()
test_file_number = len(test_id_list)
print("generating features on held-out test data...")
with tf.Session() as sess:
new_saver = tf.train.import_meta_graph(
os.path.join(self.ckpt_dir, "mymodel.ckpt.meta"))
print("loading the model parameters...")
output_layer = tf.get_collection("output_layer")[0]
input_layer = tf.get_collection("input_layer")[0]
new_saver.restore(sess, os.path.join(self.ckpt_dir,
"mymodel.ckpt"))
print("The model parameters are successfully restored")
for utt_index in range(test_file_number):
gen_test_file_name = gen_test_file_list[utt_index]
temp_test_x = test_x[test_id_list[utt_index]]
num_of_rows = temp_test_x.shape[0]
if not sequential_training:
is_training_batch = tf.get_collection(
"is_training_batch")[0]
if self.dropout_rate != 0.0:
is_training_drop = tf.get_collection(
"is_training_drop")[0]
y_predict = sess.run(output_layer,
feed_dict={
input_layer: temp_test_x,
is_training_drop: False,
is_training_batch: False
})
else:
y_predict = sess.run(output_layer,
feed_dict={
input_layer: temp_test_x,
is_training_batch: False
})
else:
temp_test_x = np.reshape(temp_test_x,
[1, num_of_rows, self.n_in])
hybrid = 0
utt_length_placeholder = tf.get_collection("utt_length")[0]
if "tanh" in self.hidden_layer_type:
hybrid = 1
is_training_batch = tf.get_collection(
"is_training_batch")[0]
if self.dropout_rate != 0.0:
is_training_drop = tf.get_collection(
"is_training_drop")[0]
if hybrid:
y_predict = sess.run(output_layer,
feed_dict={
input_layer:
temp_test_x,
utt_length_placeholder:
[num_of_rows],
is_training_drop:
False,
is_training_batch:
False
})
else:
y_predict = sess.run(output_layer,
feed_dict={
input_layer:
temp_test_x,
utt_length_placeholder:
[num_of_rows],
is_training_drop:
False
})
elif hybrid:
y_predict = sess.run(output_layer,
feed_dict={
input_layer:
temp_test_x,
utt_length_placeholder:
[num_of_rows],
is_training_batch:
False
})
else:
y_predict = sess.run(output_layer,
feed_dict={
input_layer:
temp_test_x,
utt_length_placeholder:
[num_of_rows]
})
data_utils.denorm_data(y_predict, out_scaler)
io_funcs.array_to_binary_file(y_predict, gen_test_file_name)
data_utils.drawProgressBar(utt_index + 1, test_file_number)