def infer(self, inputs_per_batch, outputs_per_batch, output_file):
# this function assumes it is run on 1 gpu with batch size of 1
with codecs.open(output_file, 'w', 'utf-8') as fout:
for step in range(len(inputs_per_batch)):
input_values = inputs_per_batch[step][0][0]
output_values = outputs_per_batch[step][0]
output_string = text_ids_to_string(
output_values[0],
self.data_layer.params['target_idx2seq'],
S_ID=self.decoder.params['GO_SYMBOL'],
EOS_ID=self.decoder.params['END_SYMBOL'],
PAD_ID=self.decoder.params['PAD_SYMBOL'],
ignore_special=True,
delim=' ',
)
input_string = text_ids_to_string(
input_values[0],
self.data_layer.params['source_idx2seq'],
S_ID=self.decoder.params['GO_SYMBOL'],
EOS_ID=self.decoder.params['END_SYMBOL'],
PAD_ID=self.decoder.params['PAD_SYMBOL'],
ignore_special=True,
delim=' ',
)
fout.write(output_string + "\n")
if step % 200 == 0:
if six.PY2:
input_string = input_string.encode('utf-8')
output_string = output_string.encode('utf-8')
deco_print("Input sequence: {}".format(input_string))
deco_print("Output sequence: {}".format(output_string))
deco_print("")
def infer(self, input_values, output_values):
input_strings, output_strings = [], []
input_values = input_values['source_tensors']
for input_sample, output_sample in zip(input_values, output_values):
output_strings.append(
text_ids_to_string(
output_sample[0],
self.get_data_layer().params['target_idx2seq'],
S_ID=self.decoder.params['GO_SYMBOL'],
EOS_ID=self.decoder.params['END_SYMBOL'],
PAD_ID=self.decoder.params['PAD_SYMBOL'],
ignore_special=True,
delim=' ',
))
input_strings.append(
text_ids_to_string(
input_sample[0],
self.get_data_layer().params['source_idx2seq'],
S_ID=self.decoder.params['GO_SYMBOL'],
EOS_ID=self.decoder.params['END_SYMBOL'],
PAD_ID=self.decoder.params['PAD_SYMBOL'],
ignore_special=True,
delim=' ',
))
return input_strings, output_strings
def infer(self, input_values, output_values):
input_strings, output_strings = [], []
input_values = input_values['source_tensors']
for input_sample, output_sample in zip(input_values, output_values):
for i in range(
0, input_sample.shape[0]): # iterate over batch dimension
output_strings.append(
text_ids_to_string(
output_sample[i],
self.get_data_layer().params['target_idx2seq'],
S_ID=self.decoder.params.get(
'GO_SYMBOL', SpecialTextTokens.S_ID.value),
EOS_ID=self.decoder.params.get(
'END_SYMBOL', SpecialTextTokens.EOS_ID),
PAD_ID=self.decoder.params.get(
'PAD_SYMBOL', SpecialTextTokens.PAD_ID),
ignore_special=True,
delim=' ',
))
input_strings.append(
text_ids_to_string(
input_sample[i],
self.get_data_layer().params['source_idx2seq'],
S_ID=self.decoder.params.get(
'GO_SYMBOL', SpecialTextTokens.S_ID.value),
EOS_ID=self.decoder.params.get(
'END_SYMBOL', SpecialTextTokens.EOS_ID.value),
PAD_ID=self.decoder.params.get(
'PAD_SYMBOL', SpecialTextTokens.PAD_ID),
ignore_special=True,
delim=' ',
))
return input_strings, output_strings
def infer(self, input_values, output_values):
input_strings, output_strings = [], []
input_values = input_values['source_tensors']
for input_sample, output_sample in zip(input_values, output_values):
output_strings.append(text_ids_to_string(
output_sample[0],
self.get_data_layer().params['target_idx2seq'],
S_ID=self.decoder.params['GO_SYMBOL'],
EOS_ID=self.decoder.params['END_SYMBOL'],
PAD_ID=self.decoder.params['PAD_SYMBOL'],
ignore_special=True, delim=' ',
))
input_strings.append(text_ids_to_string(
input_sample[0],
self.get_data_layer().params['source_idx2seq'],
S_ID=self.decoder.params['GO_SYMBOL'],
EOS_ID=self.decoder.params['END_SYMBOL'],
PAD_ID=self.decoder.params['PAD_SYMBOL'],
ignore_special=True, delim=' ',
))
return input_strings, output_strings