def get_translation_settings(self):
"""
Returns a `TranslationSettings` object based on the parsed console
arguments.
"""
args = self.parse_args()
return TranslationSettings(args)
def test_ende(self):
with open('en-de/in', 'r', encoding='utf-8') as in_file, \
open('en-de/out', 'w', encoding='utf-8') as out_file:
os.chdir('models/en-de/')
settings = TranslationSettings()
settings.input = in_file
settings.output = out_file
settings.models = ["model.npz"]
settings.beam_size = 12
settings.normalization_alpha = 1.0
settings.translation_strategy = 'sampling'
settings.sampling_temperature = 0.4
translate(settings=settings)
os.chdir('../..')
self.outputEqual('en-de/ref2','en-de/out')
def translate(self, segment):
"""
Processes a translation request.
"""
translations = self._translator.translate_string(
segment, TranslationSettings())
result = " ".join(translations[0].target_words)
return result
def get_settings(model, beam):
"""
Initialize and customize settings.
"""
translation_settings = TranslationSettings()
translation_settings.models = [model]
translation_settings.num_processes = 1
translation_settings.beam_width = beam
translation_settings.normalization_alpha = 1.0
translation_settings.suppress_unk = True
translation_settings.get_word_probs = False
return translation_settings
def get_settings():
"""
Initialize and customize settings.
"""
translation_settings = TranslationSettings()
translation_settings.models = ["models/model.npz"]
translation_settings.num_processes = 1
translation_settings.beam_width = 12
translation_settings.normalization_alpha = 1.0
translation_settings.suppress_unk = True
translation_settings.get_word_probs = True
return translation_settings
def __init__(self, request):
"""
Initialises a translation request.
@type raw_body: str
@param raw_body: the POST request submitted to Nematus server.
"""
self._request = request
self.segments = []
self.settings = TranslationSettings() # default values
self._parse()
def get_settings(self):
"""
Initialize and customize settings.
"""
translation_settings = TranslationSettings()
translation_settings.input = open('../../en-de/in')
translation_settings.output = open('../../en-de/out', 'w')
translation_settings.models = ["model.npz"]
translation_settings.beam_size = 12
translation_settings.normalization_alpha = 1.0
return translation_settings
def test_ende(self):
with open('en-de/in', 'r', encoding='utf-8') as in_file, \
open('en-de/out', 'w', encoding='utf-8') as out_file:
os.chdir('models/en-de/')
settings = TranslationSettings()
settings.input = in_file
settings.output = out_file
settings.models = ["model.npz"]
settings.beam_size = 12
settings.normalization_alpha = 1.0
translate(settings=settings)
os.chdir('../..')
self.outputEqual('en-de/ref2','en-de/out')
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)
if __name__ == "__main__":
# Parse console arguments.
settings = TranslationSettings(from_console_arguments=True)
main(settings)
def outputEqual(output1, output2):
"""given two translation outputs, check that output string is identical
"""
for i, (line, line2) in enumerate(
zip(open(output1).readlines(),
open(output2).readlines())):
#assertEqual(line.strip(), line2.strip())
print "translate {}".format(i)
print line
print line2
# English-German WMT16 system, no dropout
"""
Initialize and customize settings.
"""
translation_settings = TranslationSettings()
translation_settings.models = ["model_test/model.npz-80000"]
#translation_settings.num_processes = 8
translation_settings.beam_width = 10
translation_settings.normalization_alpha = 1.0
translation_settings.verbose = True
translation_settings.n_best = True
translation_settings.suppress_unk = True
translate(input_file=open('data/translated/fr.00'),
output_file=open('data/fr.00.8.out', 'w'),
translation_settings=translation_settings)
print "everyting ok"
#outputEqual('en-de/ref2','en-de/out')