def test_merge_colliding_spans_close_epsilon(self):
spans = [[0, 1], [1, 2], [22, 40]]
# print(sorted_spans)
res = merge_colliding_spans(spans, 20)
self.assertEqual(1, len(res))
sp = res[0]
self.assertEqual(0, sp[0])
self.assertEqual(40, sp[1])
def test_merge_colliding_spans_close(self):
spans = [[0, 1], [1, 2]]
# print(sorted_spans)
res = merge_colliding_spans(spans)
self.assertEqual(1, len(res))
sp = res[0]
self.assertEqual(0, sp[0])
self.assertEqual(2, sp[1])
def find_attributes_in_sections(self, subdoc: LegalDocumentExt,
structural_level: CharterStructuralLevel):
# finding Subjects
_subject_attentions_map = get_charter_subj_attentions(
subdoc, self.get_subj_patterns_embeddings()) # dictionary
subject_spans: Spans = collect_subjects_spans2(
subdoc, _subject_attentions_map)
# finding Values(amounts)
values: [ContractValue
] = find_value_sign_currency_attention(subdoc,
None,
absolute_spans=False)
self._rename_margin_values_tags(values)
valued_sentence_spans: Spans = collect_sentences_having_constraint_values(
subdoc, values, merge_spans=True)
_united_spans: Spans = []
for _s in valued_sentence_spans:
_united_spans.append(_s)
for _s in subject_spans:
_united_spans.append(_s)
_united_spans = merge_colliding_spans(_united_spans,
eps=-1) # TODO: check this
self.attribute_spans_to_subjects(
_united_spans,
subdoc,
structural_level # OrgStructuralLevel.BoardOfDirectors
)
# offsetting tags to absolute values
for value in values:
value += subdoc.start
for competence_tag in structural_level.competences:
competence_tag += subdoc.start
# nesting values (assigning parents)
for competence in structural_level.competences: # contract subjects
for value in values:
v_group = value.parent
if competence.contains(v_group.span):
# v_group.set_parent_tag(competence_tag)
competence.constraints.append(value.as_ContractPrice())
def collect_subjects_spans2(subdoc,
subject_attentions_map,
min_len=20) -> Spans:
spans = []
for subj in subject_attentions_map.keys():
subject_attention = subject_attentions_map[subj]
paragraph_span, confidence, _ = _find_most_relevant_paragraph(
subdoc,
subject_attention,
min_len=min_len,
return_delimiters=False)
if confidence > HyperParameters.charter_subject_attention_confidence:
if paragraph_span not in spans:
spans.append(paragraph_span)
unique_sentence_spans = merge_colliding_spans(spans, eps=-1)
return unique_sentence_spans
def collect_sentences_having_constraint_values(subdoc: LegalDocumentExt,
contract_values: [
ContractValue
],
merge_spans=True) -> Spans:
# collect sentences having constraint values
unique_sentence_spans: Spans = []
for contract_value in contract_values:
contract_value_sentence_span = subdoc.sentence_at_index(
contract_value.parent.span[0], return_delimiters=False)
if contract_value_sentence_span not in unique_sentence_spans:
unique_sentence_spans.append(contract_value_sentence_span)
contract_value_sentence_span = subdoc.sentence_at_index(
contract_value.parent.span[1], return_delimiters=False)
if contract_value_sentence_span not in unique_sentence_spans:
unique_sentence_spans.append(contract_value_sentence_span)
# --
# TODO: do not join here, join by subject
if merge_spans:
unique_sentence_spans = merge_colliding_spans(unique_sentence_spans,
eps=1)
return unique_sentence_spans
