def _create_modalities(self):
""" Creates source and target modalities.
Returns: A tuple `(input_modality, target_modality)`.
"""
input_modality_params = deep_merge_dict(
self.params["modality.params"],
self.params["modality.source.params"])
input_modality = Modality(
params=input_modality_params,
mode=self.mode,
vocab_size=self._vocab_source.vocab_size,
body_input_depth=self.params["embedding.dim.source"],
name="input_symbol_modality",
verbose=self.verbose)
target_modality_params = deep_merge_dict(
self.params["modality.params"],
self.params["modality.target.params"])
target_modality = Modality(
params=target_modality_params,
mode=self.mode,
vocab_size=self._vocab_target.vocab_size,
body_input_depth=self.params["embedding.dim.target"],
name="target_symbol_modality",
verbose=self.verbose)
return input_modality, target_modality
def main(_argv):
model_configs = maybe_load_yaml(DEFAULT_EVAL_CONFIGS)
# load flags from config file
model_configs = load_from_config_path(FLAGS.config_paths, model_configs)
# replace parameters in configs_file with tf FLAGS
model_configs = update_eval_model_configs(model_configs, FLAGS)
model_configs = deep_merge_dict(model_configs, ModelConfigs.load(FLAGS.model_dir))
model_configs = update_eval_model_configs(model_configs, FLAGS)
runner = EvalExperiment(model_configs=model_configs)
runner.run()
def main(_argv):
model_configs = maybe_load_yaml(DEFAULT_EVAL_CONFIGS)
# load flags from config file
model_configs = load_from_config_path(FLAGS.config_paths, model_configs)
# replace parameters in configs_file with tf FLAGS
model_configs = update_eval_model_configs(model_configs, FLAGS)
model_configs = deep_merge_dict(model_configs,
ModelConfigs.load(FLAGS.model_dir))
model_configs = update_eval_model_configs(model_configs, FLAGS)
runner = EvalExperiment(model_configs=model_configs)
runner.run()
def main(_argv):
# load flags from config file
model_configs = load_from_config_path(FLAGS.config_paths)
# replace parameters in configs_file with tf FLAGS
model_configs = update_configs_from_flags(model_configs, FLAGS,
EVAL_ARGS.keys())
model_configs = deep_merge_dict(model_configs,
ModelConfigs.load(FLAGS.model_dir))
model_configs = update_configs_from_flags(model_configs, FLAGS,
EVAL_ARGS.keys())
runner = EvalExperiment(model_configs=model_configs)
runner.run()
def _create_modalities(self):
""" Creates source and target modalities.
Returns: A tuple `(input_modality, target_modality)`.
"""
input_modality_params = deep_merge_dict(
self.params["modality.params"], self.params["modality.source.params"])
input_modality = Modality(
params=input_modality_params,
mode=self.mode,
vocab_size=self._vocab_source.vocab_size,
body_input_depth=self.params["embedding.dim.source"],
name="input_symbol_modality",
verbose=self.verbose)
target_modality_params = deep_merge_dict(
self.params["modality.params"], self.params["modality.target.params"])
target_modality = Modality(
params=target_modality_params,
mode=self.mode,
vocab_size=self._vocab_target.vocab_size,
body_input_depth=self.params["embedding.dim.target"],
name="target_symbol_modality",
verbose=self.verbose)
return input_modality, target_modality
def main(_argv):
model_configs = maybe_load_yaml(DEFAULT_INFER_CONFIGS)
# load flags from config file
model_configs = load_from_config_path(FLAGS.config_paths, model_configs)
# replace parameters in configs_file with tf FLAGS
model_configs = update_infer_model_configs(model_configs, FLAGS)
model_dirs = FLAGS.model_dir.strip().split(",")
if len(model_dirs) == 1:
model_configs = deep_merge_dict(model_configs, ModelConfigs.load(model_dirs[0]))
model_configs = update_infer_model_configs(model_configs, FLAGS)
runner = InferExperiment(model_configs=model_configs)
else:
runner = EnsembleExperiment(model_configs=model_configs, model_dirs=model_dirs,
weight_scheme=FLAGS.weight_scheme)
runner.run()
def main(_argv):
model_configs = maybe_load_yaml(DEFAULT_INFER_CONFIGS)
# load flags from config file
model_configs = load_from_config_path(FLAGS.config_paths, model_configs)
# replace parameters in configs_file with tf FLAGS
model_configs = update_infer_model_configs(model_configs, FLAGS)
model_dirs = FLAGS.model_dir.strip().split(",")
if len(model_dirs) == 1:
model_configs = deep_merge_dict(model_configs,
ModelConfigs.load(model_dirs[0]))
runner = InferExperiment(model_configs=model_configs)
else:
runner = EnsembleExperiment(model_configs=model_configs,
model_dirs=model_dirs,
weight_scheme=FLAGS.weight_scheme)
runner.run()