def main(settings):
"""
Translates a source language file (or STDIN) into a target language file
(or STDOUT).
"""
# Start logging.
level = logging.DEBUG if settings.verbose else logging.INFO
logging.basicConfig(level=level, format='%(levelname)s: %(message)s')
# Create the TensorFlow session.
tf_config = tf.ConfigProto()
tf_config.allow_soft_placement = True
session = tf.Session(config=tf_config)
# Load config file for each model.
configs = []
for model in settings.models:
config = load_config_from_json_file(model)
setattr(config, 'reload', model)
configs.append(config)
# Create the model graphs and restore their variables.
logging.debug("Loading models\n")
models = []
# ============= 19/8/16 KP ============
warning('='*20 + 'Model Config to Load')
warning(settings.models)
# =====================================
for i, config in enumerate(configs):
with tf.variable_scope("model%d" % i) as scope:
if config.model_type == "transformer":
model = TransformerModel(config)
else:
model = rnn_model.RNNModel(config)
saver = model_loader.init_or_restore_variables(config, session,
ensemble_scope=scope)
model.sampling_utils = SamplingUtils(settings)
models.append(model)
# ============= 19/8/16 KP ============
model_summary()
# =====================================
# TODO Ensembling is currently only supported for RNNs, so if
# TODO len(models) > 1 then check models are all rnn
# Translate the source file.
inference.translate_file(input_file=settings.input,
output_file=settings.output,
session=session,
models=models,
configs=configs,
beam_size=settings.beam_size,
nbest=settings.n_best,
minibatch_size=settings.minibatch_size,
maxibatch_size=settings.maxibatch_size,
normalization_alpha=settings.normalization_alpha)
def main(settings):
"""
Translates a source language file (or STDIN) into a target language file
(or STDOUT).
"""
# Create the TensorFlow session.
tf_config = tf.ConfigProto()
tf_config.allow_soft_placement = True
session = tf.Session(config=tf_config)
# Load config file for each model.
configs = []
for model in settings.models:
config = load_config_from_json_file(model)
setattr(config, 'reload', model)
configs.append(config)
# Create the model graphs.
logging.debug("Loading models\n")
models = []
for i, config in enumerate(configs):
with tf.variable_scope("model%d" % i) as scope:
if config.model_type == "transformer":
model = TransformerModel(config)
else:
model = rnn_model.RNNModel(config)
model.sampling_utils = SamplingUtils(settings)
models.append(model)
# Add smoothing variables (if the models were trained with smoothing).
#FIXME Assumes either all models were trained with smoothing or none were.
if configs[0].exponential_smoothing > 0.0:
smoothing = ExponentialSmoothing(configs[0].exponential_smoothing)
# Restore the model variables.
for i, config in enumerate(configs):
with tf.variable_scope("model%d" % i) as scope:
_ = model_loader.init_or_restore_variables(config, session,
ensemble_scope=scope)
# Swap-in the smoothed versions of the variables.
if configs[0].exponential_smoothing > 0.0:
session.run(fetches=smoothing.swap_ops)
# TODO Ensembling is currently only supported for RNNs, so if
# TODO len(models) > 1 then check models are all rnn
# Translate the source file.
inference.translate_file(input_file=settings.input,
output_file=settings.output,
session=session,
models=models,
configs=configs,
beam_size=settings.beam_size,
nbest=settings.n_best,
minibatch_size=settings.minibatch_size,
maxibatch_size=settings.maxibatch_size,
normalization_alpha=settings.normalization_alpha)
def main(settings):
"""
Translates a source language file (or STDIN) into a target language file
(or STDOUT).
"""
# Start logging.
level = logging.DEBUG if settings.verbose else logging.INFO
logging.basicConfig(level=level, format='%(levelname)s: %(message)s')
# Create the TensorFlow session.
if settings.cpu:
logging.info("using cpu now...")
os.environ["CUDA_VISIBLE_DEVICES"] = ""
tf_config = tf.ConfigProto(device_count={'GPU': 0})
else:
os.environ["CUDA_VISIBLE_DEVICES"] = "2"
tf_config = tf.ConfigProto()
tf_config.allow_soft_placement = True
session = tf.Session(config=tf_config)
# Load config file for each model.
configs = []
for model in settings.models:
config = util.load_config(model)
compat.fill_options(config)
config['reload'] = model
configs.append(argparse.Namespace(**config))
# Create the model graphs and restore their variables.
logging.debug("Loading models")
models = []
for i, config in enumerate(configs):
with tf.variable_scope("model%d" % i) as scope:
model = rnn_model.RNNModel(config)
saver = model_loader.init_or_restore_variables(
config, session, ensemble_scope=scope)
models.append(model)
logging.debug("Models load done.")
# Translate the source file.
inference.translate_file(input_file=settings.input,
output_file=settings.output,
session=session,
models=models,
configs=configs,
beam_size=settings.beam_size,
nbest=settings.n_best,
minibatch_size=settings.minibatch_size,
maxibatch_size=settings.maxibatch_size,
normalization_alpha=settings.normalization_alpha)
def main(settings):
"""
Translates a source language file (or STDIN) into a target language file
(or STDOUT).
"""
# Start logging.
level = logging.DEBUG if settings.verbose else logging.INFO
logging.basicConfig(level=level, format='%(levelname)s: %(message)s')
# Create the TensorFlow session.
tf_config = tf.ConfigProto()
tf_config.allow_soft_placement = True
session = tf.Session(config=tf_config)
# Load config file for each model.
configs = []
for model in settings.models:
config = load_config_from_json_file(model)
setattr(config, 'reload', model)
configs.append(config)
# Create the model graphs and restore their variables.
logging.debug("Loading models\n")
models = []
for i, config in enumerate(configs):
with tf.variable_scope("model%d" % i) as scope:
if config.model_type == "transformer":
model = TransformerModel(config)
else:
model = rnn_model.RNNModel(config)
saver = model_loader.init_or_restore_variables(config, session,
ensemble_scope=scope)
models.append(model)
# TODO Ensembling is currently only supported for RNNs, so if
# TODO len(models) > 1 then check models are all rnn
# Translate the source file.
inference.translate_file(input_file=settings.input,
output_file=settings.output,
session=session,
models=models,
configs=configs,
beam_size=settings.beam_size,
nbest=settings.n_best,
minibatch_size=settings.minibatch_size,
maxibatch_size=settings.maxibatch_size,
normalization_alpha=settings.normalization_alpha)
def validate_with_script(session, model, config):
if config.valid_script == None:
return None
logging.info('Starting external validation.')
out = tempfile.NamedTemporaryFile(mode='w')
inference.translate_file(input_file=open(config.valid_source_dataset,
encoding="UTF-8"),
output_file=out,
session=session,
models=[model],
configs=[config],
beam_size=config.beam_size,
minibatch_size=config.valid_batch_size,
normalization_alpha=config.normalization_alpha)
out.flush()
args = [config.valid_script, out.name]
proc = subprocess.Popen(args,
stdin=None,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE)
stdout_bytes, stderr_bytes = proc.communicate()
encoding = locale.getpreferredencoding()
stdout = stdout_bytes.decode(encoding=encoding)
stderr = stderr_bytes.decode(encoding=encoding)
if len(stderr) > 0:
logging.info("Validation script wrote the following to standard "
"error:\n" + stderr)
if proc.returncode != 0:
logging.warning("Validation script failed (returned exit status of "
"{}).".format(proc.returncode))
return None
try:
score = float(stdout.split()[0])
except:
logging.warning("Validation script output does not look like a score: "
"{}".format(stdout))
return None
logging.info("Validation script score: {}".format(score))
return score
def validate_with_script(session, model, config):
if config.valid_script == None:
return None
logging.info('Starting external validation.')
out = tempfile.NamedTemporaryFile(mode='w')
inference.translate_file(input_file=open(config.valid_source_dataset),
output_file=out,
session=session,
models=[model],
configs=[config],
beam_size=config.beam_size,
minibatch_size=config.valid_batch_size,
normalization_alpha=config.normalization_alpha)
out.flush()
args = [config.valid_script, out.name]
proc = subprocess.Popen(args, stdin=None, stdout=subprocess.PIPE,
stderr=subprocess.PIPE)
stdout_bytes, stderr_bytes = proc.communicate()
encoding = locale.getpreferredencoding()
stdout = stdout_bytes.decode(encoding=encoding)
stderr = stderr_bytes.decode(encoding=encoding)
if len(stderr) > 0:
logging.info("Validation script wrote the following to standard "
"error:\n" + stderr)
if proc.returncode != 0:
logging.warning("Validation script failed (returned exit status of "
"{}).".format(proc.returncode))
return None
try:
score = float(stdout.split()[0])
except:
logging.warning("Validation script output does not look like a score: "
"{}".format(stdout))
return None
logging.info("Validation script score: {}".format(score))
return score