def test_multi_aspects_per_edu(self):
self._multi_aspect_per_edu_tree()
rules_extractor = EDUTreeRulesExtractor(
only_hierarchical_relations=False)
rules = rules_extractor.extract(self.discourse_tree,
[559, 560, 561, 562, 563],
doc_id=1)
expected_rules = {
1: [
EDURelation(edu1=560,
edu2=559,
relation_type='Elaboration',
gerani=0.63),
EDURelation(edu1=561,
edu2=559,
relation_type='Elaboration',
gerani=0.53),
EDURelation(edu1=562,
edu2=559,
relation_type='same-unit',
gerani=0.3),
EDURelation(edu1=563,
edu2=559,
relation_type='same-unit',
gerani=0.2),
EDURelation(edu1=561,
edu2=560,
relation_type='Elaboration',
gerani=0.75),
EDURelation(edu1=562,
edu2=560,
relation_type='same-unit',
gerani=0.4),
EDURelation(edu1=563,
edu2=560,
relation_type='same-unit',
gerani=0.3),
EDURelation(edu1=562,
edu2=561,
relation_type='same-unit',
gerani=0.5),
EDURelation(edu1=563,
edu2=561,
relation_type='same-unit',
gerani=0.4),
EDURelation(edu1=563,
edu2=562,
relation_type='Elaboration',
gerani=0.75)
]
}
self.assertEqual(rules, expected_rules)
def test_tree_parsing_and_get_rules_all(self):
rules_extractor = EDUTreeRulesExtractor(
weight_type=['gerani'], only_hierarchical_relations=False)
rules = rules_extractor.extract(self.discourse_tree,
[513, 514, 515, 516, 517], 1)
expected_rules = {
1: [
EDURelation(edu1=514,
edu2=513,
relation_type='Elaboration',
gerani=0.8),
EDURelation(edu1=515,
edu2=513,
relation_type='same-unit',
gerani=0.42),
EDURelation(edu1=516,
edu2=513,
relation_type='same-unit',
gerani=0.33),
EDURelation(edu1=517,
edu2=513,
relation_type='same-unit',
gerani=0.25),
EDURelation(edu1=515,
edu2=514,
relation_type='same-unit',
gerani=0.5),
EDURelation(edu1=516,
edu2=514,
relation_type='same-unit',
gerani=0.42),
EDURelation(edu1=517,
edu2=514,
relation_type='same-unit',
gerani=0.33),
EDURelation(edu1=516,
edu2=515,
relation_type='Elaboration',
gerani=0.6),
EDURelation(edu1=517,
edu2=515,
relation_type='Elaboration',
gerani=0.52),
EDURelation(edu1=517,
edu2=516,
relation_type='Joint',
gerani=0.7)
]
}
self.assertEqual(rules, expected_rules)
def test_bfs_for_several_aspects_in_one_edu(self):
rules_extractor = EDUTreeRulesExtractor(
weight_type=['gerani'], only_hierarchical_relations=True)
rules = rules_extractor.extract(self.discourse_tree,
[513, 514, 515, 516, 517], 1)
expected_rules = {
1: [
EDURelation(edu1=514,
edu2=513,
relation_type='Elaboration',
gerani=0.8),
EDURelation(edu1=516,
edu2=515,
relation_type='Elaboration',
gerani=0.6),
EDURelation(edu1=517,
edu2=515,
relation_type='Elaboration',
gerani=0.52)
]
}
self.assertEqual(rules, expected_rules)
def _with_rules():
return [
EDURelation(0, 1, weight=10, relation_type="Elaboration"),
EDURelation(0, 1, weight=20, relation_type="Elaboration"),
EDURelation(0, 1, weight=11, relation_type="Contrast"),
EDURelation(0, 1, weight=22, relation_type="Contrast"),
EDURelation(2, 1, weight=10, relation_type="Elaboration"),
EDURelation(0, 1, weight=30, relation_type="Elaboration"),
]
def filter_top_n_rules(rules: List[EDURelation],
aggregation_fn: Callable = None,
top_n: int = 1) -> List[EDURelation]:
if aggregation_fn is None:
aggregation_fn = partial(max)
# sort for groupby
rules = sorted(rules, key=lambda relation: relation[:3])
rules_filtered = [
EDURelation(*(group + (aggregation_fn([r.weight
for r in relations]), )))
for group, relations in groupby(rules,
key=lambda relation: relation[:3])
]
if top_n is None:
return rules_filtered
else:
return sorted(rules, key=attrgetter("weight"), reverse=True)[:top_n]
def filter_rules_gerani(rules: List[EDURelation],
aggregation_fn: Callable = None) -> List[EDURelation]:
"""
Filter rules by its weight for each Discourse Tree.
We can have many rules per tree.
Parameters
----------
:param rules: List of relation tuples
:param aggregation_fn: Function to aggregate repeated tuple's weight.
By default it is max weight.
Returns
-------
rules_filtered : list
List of rules/relations between aspects with their maximum
gerani weights.
Examples
[
EDURelation(edu1=u'screen', edu2=u'phone',
relation_type='Elaboration', weight=1.38),
EDURelation(edu1=u'speaker', edu2=u'sound',
relation_type='Elaboration', weight=0.29),
EDURelation(edu1=u'speaker', edu2=u'sound',
relation_type='Elaboration', weight=0.21)
]
"""
if aggregation_fn is None:
aggregation_fn = partial(max)
rules = sorted(rules, key=lambda relation: relation[:3])
return [
EDURelation(*(group + (aggregation_fn([r.weight
for r in relations]), )))
for group, relations in groupby(rules,
key=lambda relation: relation[:3])
]
return [
EDURelation(0, 1, weight=10, relation_type="Elaboration"),
EDURelation(0, 1, weight=20, relation_type="Elaboration"),
EDURelation(0, 1, weight=11, relation_type="Contrast"),
EDURelation(0, 1, weight=22, relation_type="Contrast"),
EDURelation(2, 1, weight=10, relation_type="Elaboration"),
EDURelation(0, 1, weight=30, relation_type="Elaboration"),
]
@pytest.mark.parametrize(
"rules, aggregation_fn",
[
(
[
EDURelation(0, 1, weight=30, relation_type="Elaboration"),
EDURelation(0, 1, weight=22, relation_type="Contrast"),
EDURelation(2, 1, weight=10, relation_type="Elaboration"),
],
partial(max),
),
(
[
EDURelation(0, 1, weight=10, relation_type="Elaboration"),
EDURelation(0, 1, weight=11, relation_type="Contrast"),
EDURelation(2, 1, weight=10, relation_type="Elaboration"),
],
partial(min),
),
],
)