def read_model_predictions(prediction_path):
"""Read a model's validation predictions and convert to a dictionary.
Args:
prediction_path: Path to read predictions from
Returns:
A dictionary where values are predictions, and keys are composed of
the claim_id/wikipedia_url/sentence_id/scrape_type
"""
model_predictions = util.read_json(prediction_path)
id_to_predictions = {}
for pred in model_predictions['predictions']:
claim_id = pred['metadata']['claim_id']
scrape_type = pred['metadata']['scrape_type']
wikipedia_url = pred['metadata']['wikipedia_url']
sentence_id = pred['metadata']['sentence_id']
identifier = make_example_id(
claim_id=claim_id,
wikipedia_url=wikipedia_url,
sentence_id=sentence_id,
scrape_type=scrape_type,
)
id_to_predictions[identifier] = pred
return id_to_predictions
def load_from_file(cls, filename_prefix):
conf = util.read_json(filename_prefix + '.space_tokenizer')
return cls(**conf)
def load_from_file(cls, filename_prefix):
params = util.read_json(f'{filename_prefix}.tokenizer')
bert_tokenizer = BertTokenizer(vocab_file=params['vocab_file'],
do_lower_case=params['do_lower_case'])
return bert_tokenizer
def load_from_file(cls, filename_prefix):
conf = util.read_json(filename_prefix + '.regex_tokenizer')
tokenizer_cls = conf['tokenizer_cls']
tokenizer = tokenizer_registry[tokenizer_cls].load_from_file(
filename_prefix)
return cls(tokenizer=tokenizer, reserved_re=conf['reserved_re'])
