def __init__(self,
types_path: str,
tokenizer: BertTokenizer,
neg_entity_count: int = None,
neg_rel_count: int = None,
max_span_size: int = None,
logger: Logger = None,
**kwargs):
types = json.load(
open(types_path),
object_pairs_hook=OrderedDict) # entity + relation types
self._entity_types = OrderedDict()
self._idx2entity_type = OrderedDict()
self._relation_types = OrderedDict()
self._idx2relation_type = OrderedDict()
# entities
# add 'None' entity type
none_entity_type = EntityType('None', 0, 'None', 'No Entity')
self._entity_types['None'] = none_entity_type
self._idx2entity_type[0] = none_entity_type
# specified entity types
for i, (key, v) in enumerate(types['entities'].items()):
entity_type = EntityType(key, i + 1, v['short'], v['verbose'])
self._entity_types[key] = entity_type
self._idx2entity_type[i + 1] = entity_type
# relations
# add 'None' relation type
none_relation_type = RelationType('None', 0, 'None', 'No Relation')
self._relation_types['None'] = none_relation_type
self._idx2relation_type[0] = none_relation_type
# specified relation types
for i, (key, v) in enumerate(types['relations'].items()):
relation_type = RelationType(key, i + 1, v['short'], v['verbose'],
v['symmetric'])
self._relation_types[key] = relation_type
self._idx2relation_type[i + 1] = relation_type
self._neg_entity_count = neg_entity_count
self._neg_rel_count = neg_rel_count
self._max_span_size = max_span_size
self._datasets = dict()
self._tokenizer = tokenizer
self._logger = logger
self._vocabulary_size = tokenizer.vocab_size
def __init__(self, dataset: Dataset, input_reader: BaseInputReader,
text_encoder: BertTokenizer, rel_filter_threshold: float,
no_overlapping: bool, predictions_path: str,
examples_path: str, example_count: int):
self._text_encoder = text_encoder
self._input_reader = input_reader
self._dataset = dataset
self._rel_filter_threshold = rel_filter_threshold
self._no_overlapping = no_overlapping
self._predictions_path = predictions_path
self._examples_path = examples_path
self._example_count = example_count
# relations
self._gt_relations = [] # ground truth
self._pred_relations = [] # prediction
# entities
self._gt_entities = [] # ground truth
self._pred_entities = [] # prediction
self._pseudo_entity_type = EntityType(
'Entity', 1, 'Entity', 'Entity') # for span only evaluation
self._convert_gt(self._dataset.documents)
def __init__(self, dataset: Dataset, input_reader: JsonInputReader, text_encoder: BertTokenizer,
rel_filter_threshold: float, example_count: int, example_path: str,
epoch: int, dataset_label: str):
self._text_encoder = text_encoder
self._input_reader = input_reader
self._dataset = dataset
self._rel_filter_threshold = rel_filter_threshold
self._epoch = epoch
self._dataset_label = dataset_label
self._example_count = example_count
self._examples_path = example_path
# relations
self._gt_relations = [] # ground truth
self._pred_relations = [] # prediction
# entities
self._gt_entities = [] # ground truth
self._pred_entities = [] # prediction
self._pseudo_entity_type = EntityType('Entity', 1, 'Entity', 'Entity') # for span only evaluation
self._convert_gt(self._dataset.documents)
def __init__(self, pred_dataset: Dataset, dataset: Dataset):
# super(CustomEvaluator, self).__init__(dataset, input_reader, text_encoder,
# rel_filter_threshold, no_overlapping,
# predictions_path, examples_path, example_count)
self._gt_entities = []
self._gt_relations = []
self._pred_entities = []
self._pred_relations = []
self._pseudo_entity_type = EntityType('Entity', 1, 'Entity', 'Entity')
self._convert_gt(dataset.documents)
self._convert_gt(pred_dataset.documents, val=False)