def entrance(mode, config, estimator=None):
if config is not None:
Config.load_config(config)
if Config.estimator_name == 'esim':
main_fasttext(mode, config, estimator)
else:
main(mode, config, estimator)
def main(args=NullArgs()):
LogHelper.setup()
logger = LogHelper.get_logger(
os.path.splitext(os.path.basename(__file__))[0])
args.mode = Mode.PREDICT
if args.config is not None:
Config.load_config(args.config)
if args.out_file is not None:
Config.relative_path_submission = args.out_file
if args.in_file is not None:
Config.relative_path_test_file = args.in_file
if args.database is not None:
Config.relative_path_db = args.database
print("relative_path_db " + Config.relative_path_db)
print("raw_test_set " + Config.raw_test_set())
if os.path.exists(Config.test_doc_file):
os.remove(Config.test_doc_file)
if os.path.exists(Config.test_set_file):
os.remove(Config.test_set_file)
if args.mode in {Mode.PIPELINE, Mode.PREDICT, Mode.PREDICT_ALL_DATASETS}:
logger.info(
"=========================== Sub-task 1. Document Retrieval =========================================="
)
document_retrieval(logger, args.mode)
if args.mode in {
Mode.PIPELINE_NO_DOC_RETR, Mode.PIPELINE, Mode.PREDICT,
Mode.PREDICT_NO_DOC_RETR, Mode.PREDICT_ALL_DATASETS,
Mode.PREDICT_NO_DOC_RETR_ALL_DATASETS
}:
logger.info(
"=========================== Sub-task 2. Sentence Retrieval =========================================="
)
sentence_retrieval_ensemble(logger, args.mode)
logger.info(
"=========================== Sub-task 3. Claim Validation ============================================"
)
rte(logger, args, args.mode)
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--mode', help='\'train\' or \'test\'', required=True)
parser.add_argument('--config',
help='/path/to/config/file, in JSON format')
args = parser.parse_args()
print(args)
print(type(args))
LogHelper.setup()
logger = LogHelper.get_logger(
os.path.splitext(os.path.basename(__file__))[0] + "_" + args.mode)
logger.info("parameters:\n" + str(vars(args)))
if args.config is not None:
Config.load_config(args.config)
logger.info("scorer type: " + Config.estimator_name)
logger.info("random seed: " + str(Config.seed))
logger.info("ESIM arguments: " + str(Config.esim_hyper_param))
# loading FastText takes long time, so better pickle the loaded FastText model
# if os.path.splitext(Config.fasttext_path)[1] == '.p':
# with open(Config.fasttext_path, "rb") as ft_file:
# fasttext_model = pickle.load(ft_file)
# else:
# fasttext_model = Config.fasttext_path
if hasattr(Config, 'is_snopes'):
is_snopes = Config.is_snopes
else:
is_snopes = False
logger.debug("is_snopes: " + str(is_snopes))
def main(mode: RTERunPhase, config=None, estimator=None):
LogHelper.setup()
logger = LogHelper.get_logger(
os.path.splitext(os.path.basename(__file__))[0] + "_" + str(mode))
if config is not None and isinstance(config, str):
logger.info("model: " + str(mode) + ", config: " + str(config))
Config.load_config(config)
logger.info("scorer type: " + Config.estimator_name)
logger.info("random seed: " + str(Config.seed))
logger.info("ESIM arguments: " + str(Config.esim_hyper_param))
# loading FastText takes long time, so better pickle the loaded FastText model
if os.path.splitext(Config.fasttext_path)[1] == '.p':
with open(Config.fasttext_path, "rb") as ft_file:
fasttext_model = pickle.load(ft_file)
else:
fasttext_model = Config.fasttext_path
if mode == RTERunPhase.train:
# # training mode
training_set, fasttext_model, vocab, embeddings = embed_data_set_with_glove_and_fasttext_claim_only(
Config.training_set_file,
fasttext_model,
glove_path=Config.glove_path,
threshold_h_sent_size=Config.max_sentence_size)
h_sent_sizes = training_set['data']['h_sent_sizes']
h_sizes = np.ones(len(h_sent_sizes), np.int32)
training_set['data']['h_sent_sizes'] = np.expand_dims(h_sent_sizes, 1)
training_set['data']['h_sizes'] = h_sizes
training_set['data']['h_np'] = np.expand_dims(
training_set['data']['h_np'], 1)
training_set['data']['h_ft_np'] = np.expand_dims(
training_set['data']['h_ft_np'], 1)
valid_set, _, _, _ = embed_data_set_with_glove_and_fasttext_claim_only(
Config.dev_set_file,
fasttext_model,
vocab_dict=vocab,
glove_embeddings=embeddings,
threshold_h_sent_size=Config.max_sentence_size)
del fasttext_model
h_sent_sizes = valid_set['data']['h_sent_sizes']
h_sizes = np.ones(len(h_sent_sizes), np.int32)
valid_set['data']['h_sent_sizes'] = np.expand_dims(h_sent_sizes, 1)
valid_set['data']['h_sizes'] = h_sizes
valid_set['data']['h_np'] = np.expand_dims(valid_set['data']['h_np'],
1)
valid_set['data']['h_ft_np'] = np.expand_dims(
valid_set['data']['h_ft_np'], 1)
X_dict = {
'X_train': training_set['data'],
'X_valid': valid_set['data'],
'y_valid': valid_set['label'],
'embedding': embeddings
}
if estimator is None:
estimator = get_estimator(Config.estimator_name,
Config.ckpt_folder)
if 'CUDA_VISIBLE_DEVICES' not in os.environ or not str(
os.environ['CUDA_VISIBLE_DEVICES']).strip():
os.environ['CUDA_VISIBLE_DEVICES'] = str(
GPUtil.getFirstAvailable(maxLoad=1.0,
maxMemory=1.0 -
Config.max_gpu_memory)[0])
estimator.fit(X_dict, training_set['label'])
save_model(estimator, Config.model_folder, Config.pickle_name, logger)
elif mode == 'test':
# testing mode
restore_param_required = estimator is None
if estimator is None:
estimator = load_model(Config.model_folder, Config.pickle_name)
if estimator is None:
estimator = get_estimator(Config.estimator_name,
Config.ckpt_folder)
vocab, embeddings = load_whole_glove(Config.glove_path)
vocab = vocab_map(vocab)
test_set, _, _, _ = embed_data_set_with_glove_and_fasttext_claim_only(
Config.test_set_file,
fasttext_model,
vocab_dict=vocab,
glove_embeddings=embeddings,
threshold_h_sent_size=Config.max_sentence_size)
del fasttext_model
h_sent_sizes = test_set['data']['h_sent_sizes']
h_sizes = np.ones(len(h_sent_sizes), np.int32)
test_set['data']['h_sent_sizes'] = np.expand_dims(h_sent_sizes, 1)
test_set['data']['h_sizes'] = h_sizes
test_set['data']['h_np'] = np.expand_dims(test_set['data']['h_np'], 1)
test_set['data']['h_ft_np'] = np.expand_dims(
test_set['data']['h_ft_np'], 1)
x_dict = {'X_test': test_set['data'], 'embedding': embeddings}
if 'CUDA_VISIBLE_DEVICES' not in os.environ or not str(
os.environ['CUDA_VISIBLE_DEVICES']).strip():
os.environ['CUDA_VISIBLE_DEVICES'] = str(
GPUtil.getFirstAvailable(maxLoad=1.0,
maxMemory=1.0 -
Config.max_gpu_memory)[0])
predictions = estimator.predict(x_dict, restore_param_required)
generate_submission(predictions, test_set['id'], Config.test_set_file,
Config.submission_file)
if 'label' in test_set:
print_metrics(test_set['label'], predictions, logger)
return estimator