def __run():
# Setup default evaluator.
evaluator = ThreeClassEvaluator(DataType.Test)
experiment_data = RuSentRelTrainingData(
labels_scaler=labels_scaler,
stemmer=stemmer,
evaluator=evaluator,
opinion_formatter=RuSentRelOpinionCollectionFormatter(),
callback=CallbackEvalF1NPU(DataType.Test))
extra_name_suffix = Common.create_exp_name_suffix(
use_balancing=balanced_input,
terms_per_context=terms_per_context,
dist_in_terms_between_att_ends=dist_in_terms_between_attitude_ends)
# Composing experiment.
experiment = create_experiment(exp_type=exp_type,
experiment_data=experiment_data,
folding_type=folding_type,
rusentrel_version=rusentrel_version,
experiment_io_type=CustomNetworkExperimentIO,
ruattitudes_version=ra_version,
load_ruattitude_docs=False,
extra_name_suffix=extra_name_suffix)
full_model_name = Common.create_full_model_name(folding_type=folding_type,
model_name=model_name,
input_type=model_input_type)
model_io = NeuralNetworkModelIO(
full_model_name=full_model_name,
target_dir=experiment.ExperimentIO.get_target_dir(),
# From this depends on whether we have a specific dir or not.
source_dir=None if model_name_tag is None else u"",
model_name_tag=ModelNameTagArg.NO_TAG if model_name_tag is None else model_name_tag)
# Setup model io.
experiment_data.set_model_io(model_io)
# Check dir existence in advance.
model_dir = model_io.get_model_dir()
if not exists(model_dir):
print u"Skipping [path not exists]: {}".format(model_dir)
return
engine = ExperimentF1pnuEvaluator(experiment=experiment,
data_type=DataType.Test,
max_epochs_count=max_epochs_count,
forced=force_eval)
# Starting evaluation process.
engine.run()
use_balancing=balanced_input,
terms_per_context=terms_per_context,
dist_in_terms_between_att_ends=dist_in_terms_between_attitude_ends)
experiment = create_experiment(
exp_type=exp_type,
experiment_data=experiment_data,
folding_type=folding_type,
rusentrel_version=rusentrel_version,
ruattitudes_version=ra_version,
experiment_io_type=CustomNetworkExperimentIO,
extra_name_suffix=extra_name_suffix,
load_ruattitude_docs=False)
full_model_name = Common.create_full_model_name(
folding_type=folding_type,
model_name=model_name,
input_type=model_input_type)
model_io = NeuralNetworkModelIO(
full_model_name=full_model_name,
target_dir=experiment.ExperimentIO.get_target_dir(),
source_dir=model_load_dir,
embedding_filepath=embedding_filepath,
vocab_filepath=vocab_filepath,
model_name_tag=model_name_tag)
# Setup logging dir.
callback.set_log_dir(join(model_io.get_model_dir(), Common.log_dir))
# Setup model io.
experiment_data.set_model_io(model_io)
log_formatter = logging.Formatter(
'%(asctime)s %(levelname)8s %(name)s | %(message)s')
stream_handler.setFormatter(log_formatter)
logger = logging.getLogger(__name__)
logger.setLevel(logging.INFO)
logger.addHandler(stream_handler)
# Initialize entity formatter.
entity_formatter = create_entity_formatter(
fmt_type=entity_formatter_type,
create_russian_pos_tagger_func=lambda: POSMystemWrapper(
mystem=stemmer.MystemInstance))
# Setup model name.
full_model_name = Common.create_full_model_name(
sample_fmt_type=sample_formatter_type,
entities_fmt_type=entity_formatter_type,
labels_count=labels_count)
model_io = BertModelIO(full_model_name=full_model_name)
# Create experiment data and all the related information.
experiment_data = CustomSerializationData(
labels_scaler=create_labels_scaler(labels_count),
stemmer=stemmer,
frames_version=frames_version if parse_frames else None,
model_io=model_io,
terms_per_context=terms_per_context,
dist_in_terms_between_attitude_ends=dist_in_terms_between_attitude_ends
)
extra_name_suffix = create_exp_name_suffix(
def _create_model_dir(self, folding_type, model_name, exp_type):
return Common.create_full_model_name(folding_type=folding_type,
input_type=self.__input_type,
model_name=model_name)
def _create_model_name(labels_count, sample_fmt_type):
return Common.create_full_model_name(
sample_fmt_type=sample_fmt_type,
labels_count=labels_count,
entities_fmt_type=EntityFormatterTypes.SimpleSharpPrefixed)
