def translate_lines(
output_handler: sockeye.output_handler.OutputHandler,
source_data: Iterable[str],
translator: sockeye.inference.Translator) -> Tuple[int, float]:
"""
Translates each line from source_data, calling output handler for each result.
:param output_handler: A handler that will be called once with the output of each translation.
:param source_data: A enumerable list of source sentences that will be translated.
:param translator: The translator that will be used for each line of input.
:param edge_vocab: Edge label vocabulary for graphs.
:return: The number of lines translated, and the total time taken.
"""
i = 0
total_time = 0.0
for i, line in enumerate(source_data, 1):
tic = time.time()
#########
# GCN - This is an ugly hack: we concatenate the surface sentence
# and the graph into a single line
surface, graph = line.split('\t')
#########
trans_input = translator.make_input(i, surface, graph,
translator.vocab_edge)
logger.debug(" IN: %s", trans_input)
trans_output = translator.translate(trans_input)
trans_wall_time = time.time() - tic
total_time += trans_wall_time
logger.debug("OUT: %s", trans_output)
logger.debug("OUT: time=%.2f", trans_wall_time)
output_handler.handle(trans_input, trans_output, trans_wall_time)
return i, total_time
def translate_lines(
output_handler: sockeye.output_handler.OutputHandler,
source_data: Iterable[str],
translator: sockeye.inference.Translator) -> Tuple[int, float]:
"""
Translates each line from source_data, calling output handler for each result.
:param output_handler: A handler that will be called once with the output of each translation.
:param source_data: A enumerable list of source sentences that will be translated.
:param translator: The translator that will be used for each line of input.
:return: The number of lines translated, and the total time taken.
"""
i = 0
total_time = 0.0
for i, line in enumerate(source_data, 1):
trans_input = translator.make_input(i, line)
logger.debug(" IN: %s", trans_input)
tic = time.time()
trans_output = translator.translate(trans_input)
trans_wall_time = time.time() - tic
total_time += trans_wall_time
logger.debug("OUT: %s", trans_output)
logger.debug("OUT: time=%.2f", trans_wall_time)
output_handler.handle(trans_input, trans_output)
return i, total_time
def translate(output_handler: sockeye.output_handler.OutputHandler,
source_data: Iterable[str],
translator: sockeye.inference.Translator,
chunk_id: int = 0) -> float:
"""
Translates each line from source_data, calling output handler after translating a batch.
:param output_handler: A handler that will be called once with the output of each translation.
:param source_data: A enumerable list of source sentences that will be translated.
:param translator: The translator that will be used for each line of input.
:param chunk_id: Global id of the chunk.
:return: Total time taken.
"""
tic = time.time()
trans_inputs = [
translator.make_input(i, line)
for i, line in enumerate(source_data, chunk_id + 1)
]
trans_outputs = translator.translate(trans_inputs)
total_time = time.time() - tic
batch_time = total_time / len(trans_inputs)
for trans_input, trans_output in zip(trans_inputs, trans_outputs):
output_handler.handle(trans_input, trans_output, batch_time)
return total_time