def edge2pattern(edge, root=False, subtype=False):
if root and edge.is_atom():
root_str = edge.root()
else:
root_str = '*'
if subtype:
et = edge.type()
else:
et = edge.type()[0]
pattern = '{}/{}'.format(root_str, et)
ar = edge.argroles()
if ar == '':
return hedge(pattern)
else:
return hedge('{}.{}'.format(pattern, ar))
def normalize_edge(edge):
if edge.is_atom():
return edge
conn = edge[0]
ar = conn.argroles()
if ar != '':
roles_edges = zip(ar, edge[1:])
roles_edges = sorted(roles_edges,
key=lambda role_edge: argrole_order[role_edge[0]])
ar = ''.join([role_edge[0] for role_edge in roles_edges])
pred = conn.atom()
new_pred = hedge('{}/{}.{}'.format(pred.root(), pred.type(), ar))
conn = conn.replace_atom(pred, new_pred)
edge = hedge([conn] + list(role_edge[1] for role_edge in roles_edges))
return hedge([normalize_edge(subedge) for subedge in edge])
def parse_and_add(self, text, hg, sequence=None, set_text=True):
parse_results = self.parse(text)
for parse in parse_results['parses']:
main_edge = parse['resolved_corefs']
if self.corefs:
unresolved_edge = parse['main_edge']
else:
unresolved_edge = None
# add main edge
if main_edge:
if sequence:
hg.add_to_sequence(sequence, main_edge)
else:
hg.add(main_edge)
# attach text to edge and subedges
_set_edges_text(main_edge, hg, parse)
if self.corefs:
if unresolved_edge != main_edge:
_set_edges_text(main_edge, hg, parse)
coref_res_edge = hedge(
(const.coref_res_pred, unresolved_edge, main_edge))
hg.add(coref_res_edge)
# add extra edges
for edge in parse['extra_edges']:
hg.add(edge)
for edge in parse_results['inferred_edges']:
hg.add(edge, count=True)
return parse_results
def test_add_count(self):
self.hg.destroy()
edge = hedge('(is/Pd graphbrain/Cp great/C)')
self.hg.add(edge, count=True)
self.assertEqual(self.hg.get_int_attribute(edge, 'count'), 1)
self.hg.add(edge, count=True)
self.assertEqual(self.hg.get_int_attribute(edge, 'count'), 2)
def test_parser(args):
parser = create_parser(lang=args.lang, parser_class=args.parser)
total = 0
wrong = 0
sentence = None
with open(args.infile) as f:
for line in f:
if sentence:
total += 1
correct_edge = hedge(line.strip())
parser_output = parser.parse(sentence)
parsed_sentence = parser_output['parses'][0]
edge = parsed_sentence['main_edge']
sent = parsed_sentence['spacy_sentence']
if edge != correct_edge:
wrong += 1
print_tree(sent.root)
print('expected:')
print(correct_edge)
print('result:')
print(edge)
sentence = None
else:
sentence = line.strip()
print('%s wrong out of %s.' % (wrong, total))
def input_defects(sentence, edge):
s = colored('s', 'magenta')
h = colored('h', 'cyan')
i = colored('i', 'yellow')
options_str = '{}/{}/{}/subedge'.format(s, h, i)
input_msg = 'wrong subedge ({}) ? '.format(options_str)
defects = None
while not defects:
answer = input(input_msg)
if answer == 's':
print('\n{}\n'.format(sentence))
elif answer == 'h':
print('\n{}\n'.format(colored_edge(edge)))
elif answer == 'i':
print('\n{}\n'.format(indented(edge)))
else:
edge_strs = answer.split('&')
subedges = []
failed = False
for edge_str in edge_strs:
subedge = hedge(edge_str)
if subedge is None:
error_msg('{} did not parse correctly.'.format(edge_str))
failed = True
elif edge.contains(subedge, deep=True):
subedges.append(subedge)
else:
error_msg('{} is not a subedge of {}.'.format(
subedge.to_str(), edge.to_str()))
failed = True
if not failed:
defects = subedges
return defects
def process_edge(self, edge, depth):
hg = self.system.get_hg(self)
if not edge.is_atom():
ct = edge.connector_type()
if ct[:2] == 'Pd':
pred = edge[0]
if (len(edge) > 2 and deep_lemma(hg, pred).root()
in CONFLICT_PRED_LEMMAS):
subjects = edge.edges_with_argrole('s')
objects = edge.edges_with_argrole('o')
if len(subjects) == 1 and len(objects) == 1:
subject = strip_concept(subjects[0])
obj = strip_concept(objects[0])
if (subject and obj and has_proper_concept(subject)
and has_proper_concept(obj)):
actor_orig = main_coref(hg, subject)
actor_targ = main_coref(hg, obj)
conflict_edge = hedge(
('conflict/P/.', actor_orig, actor_targ, edge))
if (is_actor(hg, actor_orig)
and is_actor(hg, actor_targ)):
yield create_op(conflict_edge)
for wedge in self._topics(
hg, actor_orig, actor_targ, edge):
yield wedge
self.conflicts += 1
def test_degrees(self):
self.hg.destroy()
graphbrain = hedge('graphbrain/1')
great = hedge('great/1')
self.assertEqual(self.hg.degree(graphbrain), 0)
self.hg.add('(is graphbrain/1 great/1)')
self.assertEqual(self.hg.degree(graphbrain), 1)
self.assertEqual(self.hg.degree(great), 1)
self.hg.add('(size graphbrain/1 7)')
self.assertEqual(self.hg.degree(graphbrain), 2)
self.assertEqual(self.hg.degree(great), 1)
self.hg.remove(hedge('(is graphbrain/1 great/1)'))
self.assertEqual(self.hg.degree(graphbrain), 1)
self.assertEqual(self.hg.degree(great), 0)
self.hg.remove(hedge('(size graphbrain/1 7)'))
self.assertEqual(self.hg.degree(graphbrain), 0)
def test_non_primary_to_primary(self):
self.hg.destroy()
edge = hedge('(is/P (the/M sun/C) shining/C)')
self.hg.add(edge, primary=False)
self.assertFalse(self.hg.is_primary(edge))
self.hg.add(edge, primary=True)
self.assertTrue(self.hg.is_primary(edge))
def clean_edge(edge):
if not edge.is_atom():
return edge
catom = edge.root()
catom = catom.replace('_', '')
catom = unidecode(catom)
return hedge(catom)
def test_counter4(self):
pc = PatternCounter(match_roots={'./P'})
pc.count(hedge('((not/M is/P.sc) mary/C (not/M nice/C))'))
self.assertTrue(pc.patterns[hedge('(*/P.sc */C */C)')] == 0)
self.assertTrue(pc.patterns[hedge('((*/M */P.sc) */C */C)')] == 0)
self.assertTrue(pc.patterns[hedge('((*/M is/P.sc) */C */C)')] == 1)
self.assertTrue(pc.patterns[hedge('(*/P.sc */C (*/M */C))')] == 0)
self.assertTrue(
pc.patterns[hedge('((*/M */P.sc) */C (*/M */C))')] == 0)
self.assertTrue(pc.patterns[hedge('(*/M */P.sc)')] == 0)
self.assertTrue(pc.patterns[hedge('(*/M is/P.sc)')] == 1)
self.assertTrue(pc.patterns[hedge('(*/M */C)')] == 1)
def replace_subject(edge, new_subject):
connector = edge[0]
new_edge = list(edge)
for pos, role in enumerate(connector.argroles()):
if role == 's':
new_edge[pos + 1] = new_subject
return hedge(new_edge)
def count(self, edge):
if not edge.is_atom():
if self._matches_expansions(edge):
for pattern in self._edge2patterns(edge):
self.patterns[hedge(pattern)] += 1
if self.count_subedges:
for subedge in edge:
self.count(subedge)
def test_counters3_non_deep_removal(self):
self.hg.destroy()
self.hg.add('(says mary/C (is graphbrain/C great/C))')
self.hg.remove(hedge('(says mary/C (is graphbrain/C great/C))'),
deep=False)
self.assertEqual(self.hg.atom_count(), 5)
self.assertEqual(self.hg.primary_atom_count(), 0)
self.assertEqual(self.hg.edge_count(), 6)
self.assertEqual(self.hg.primary_edge_count(), 0)
def blocks(edge, subtypes=False, argroles=True, namespaces=False):
edge = hedge(edge)
sedge = edge.simplify(subtypes=subtypes,
argroles=argroles,
namespaces=namespaces)
html = _edge2html_blocks(sedge)
html = '<div style="background-color:#fcfcfc; padding:50px">{}'\
'</div>'.format(html)
display(HTML(html))
def _edge2patterns(self, edge):
force_subtypes = self._force_subtypes(edge)
force_root, _ = self._force_root_expansion(edge)
return list(
hedge(pattern)
for pattern in self._list2patterns(list(edge.normalized()),
force_subtypes=force_subtypes,
force_root=force_root,
force_expansion=False))
def test_search(self):
self.hg.destroy()
self.hg.add('(is/Pd graphbrain/Cp great/C)')
self.hg.add('(says/Pd mary/Cp)')
self.hg.add('(says/Pd mary/Cp (is/Pd graphbrain/Cp great/C))')
self.hg.add('(says/Pd mary/Cp (is/Pd graphbrain/Cp great/C) extra/C)')
self.assertEqual(list(self.hg.search('(* graphbrain/Cp *)')),
[hedge('(is/Pd graphbrain/Cp great/C)')])
self.assertEqual(list(self.hg.search('(is/Pd graphbrain/Cp *)')),
[hedge('(is/Pd graphbrain/Cp great/C)')])
self.assertEqual(list(self.hg.search('(x * *)')), [])
self.assertEqual(
list(self.hg.search('(says/Pd * '
'(is/Pd graphbrain/Cp great/C))')),
[hedge('(says/Pd mary/Cp (is/Pd graphbrain/Cp great/C))')])
self.assertEqual(
list(self.hg.search('(says/Pd * (is/Pd * *))')),
[hedge('(says/Pd mary/Cp (is/Pd graphbrain/Cp great/C))')])
def test_inc_attributes_does_not_exist(self):
self.hg.destroy()
edge = hedge('(is graphbrain/1 great/1)')
self.hg.add(edge)
self.assertEqual(self.hg.get_int_attribute(edge, 'foo'), None)
self.hg.inc_attribute(edge, 'foo')
self.assertEqual(self.hg.get_int_attribute(edge, 'foo'), 1)
self.hg.inc_attribute(edge, 'foo')
self.assertEqual(self.hg.get_int_attribute(edge, 'foo'), 2)
def test_batch_adds(self):
self.hg.destroy()
edges = []
for i in range(10):
edges.append(hedge('(is/P {}/C number/C)'.format(i)))
with hopen('test.db') as hg:
for edge in edges:
hg.add(edge)
for edge in edges:
self.assertTrue(self.hg.exists(edge))
def test_counter1(self):
pc = PatternCounter()
pc.count(hedge('((not/M is/P.sc) mary/C (not/M nice/C))'))
self.assertTrue(pc.patterns[hedge('(*/P.sc */C */C)')] == 1)
self.assertTrue(pc.patterns[hedge('((*/M */P.sc) */C */C)')] == 1)
self.assertTrue(pc.patterns[hedge('(*/P.sc */C (*/M */C))')] == 1)
self.assertTrue(
pc.patterns[hedge('((*/M */P.sc) */C (*/M */C))')] == 1)
self.assertTrue(pc.patterns[hedge('(*/M */P.sc)')] == 1)
self.assertFalse(pc.patterns[hedge('(*/M is/P.sc)')] == 1)
self.assertTrue(pc.patterns[hedge('(*/M */C)')] == 1)
def show(edge, style='indented'):
"""Displays a representation of the edge in the notebook.
Keyword arguments:
style -- render style ('indented', 'line')
(default: 'indented')
"""
edge = hedge(edge)
html = _edge2html_show(edge, style=style)[0]
display(HTML(html))
def test_star_limit(self):
self.hg.destroy()
self.hg.add('(is graphbrain/1 great/1)')
self.hg.add('(is graphbrain/1 great/2)')
self.hg.add('(is graphbrain/1 great/3)')
center = hedge('graphbrain/1')
self.assertEqual(len(list(self.hg.star(center))), 3)
self.assertEqual(len(list(self.hg.star(center, limit=1))), 1)
self.assertEqual(len(list(self.hg.star(center, limit=2))), 2)
self.assertEqual(len(list(self.hg.star(center, limit=10))), 3)
def coref_set(hg, edge, corefs=None):
"""Returns the set of coreferences that the given edge belongs to."""
if corefs is None:
corefs = {edge}
for coref_edge in hg.edges_with_edges((hedge(coref_pred), edge)):
if len(coref_edge) == 3 and coref_edge[0].to_str() == coref_pred:
for item in coref_edge[1:]:
if item not in corefs:
corefs.add(item)
coref_set(hg, item, corefs)
return corefs
def test_deep_degrees(self):
self.hg.destroy()
edge1 = hedge('((is/M going/P) mary/C (to (the/M gym/C)))')
self.hg.add(edge1)
mary = hedge('mary/C')
gym = hedge('gym/C')
is_going = hedge('(is/M going/P)')
self.assertEqual(self.hg.deep_degree(edge1), 0)
self.assertEqual(self.hg.degree(mary), 1)
self.assertEqual(self.hg.deep_degree(mary), 1)
self.assertEqual(self.hg.degree(gym), 0)
self.assertEqual(self.hg.deep_degree(gym), 1)
self.assertEqual(self.hg.degree(is_going), 1)
self.assertEqual(self.hg.deep_degree(is_going), 1)
self.assertEqual(self.hg.deep_degree(gym), 1)
edge2 = hedge('((is/M going/P) john/C (to (the/M gym/C)))')
self.hg.add(edge2)
self.assertEqual(self.hg.degree(gym), 0)
self.assertEqual(self.hg.deep_degree(gym), 2)
self.assertEqual(self.hg.degree(is_going), 2)
self.assertEqual(self.hg.deep_degree(is_going), 2)
def test_add_with_attributes2(self):
self.hg.destroy()
edge = hedge('(is graphbrain/1 great/1)')
self.hg.add_with_attributes(edge, {
'p': 0,
'd': 10,
'dd': 20,
'foo': 777,
'bar': -.77,
'xx': 'def'
})
self.assertFalse(self.hg.is_primary(edge))
def test_search_star(self):
self.hg.destroy()
self.hg.add('(is/Pd graphbrain/Cp great/C)')
self.hg.add('(says/Pd mary/Cp)')
self.hg.add('(says/Pd mary/Cp (is/Pd graphbrain/Cp great/C))')
self.hg.add('(says/Pd mary/Cp (is/Pd graphbrain/Cp great/C) extra/C)')
self.assertEqual(
set(self.hg.search('*')), {
hedge('(is/Pd graphbrain/Cp great/C)'),
hedge('(says/Pd mary/Cp)'),
hedge('(says/Pd mary/Cp '
'(is/Pd graphbrain/Cp great/C))'),
hedge('(says/Pd mary/Cp '
'(is/Pd graphbrain/Cp great/C) extra/C)'),
hedge('extra/C'),
hedge('graphbrain/Cp'),
hedge('great/C'),
hedge('is/Pd'),
hedge('mary/Cp'),
hedge('says/Pd')
})
def test_counter6(self):
pc = PatternCounter(count_subedges=False, match_subtypes={'*/M'})
pc.count(hedge('((not/Mn is/Pd.sc) mary/Cp.s (very/M nice/Cc.s))'))
self.assertTrue(pc.patterns[hedge('(*/P.sc */C */C)')] == 1)
self.assertTrue(pc.patterns[hedge('((*/Mn */P.sc) */C */C)')] == 1)
self.assertTrue(pc.patterns[hedge('(*/P.sc */C (*/M */C))')] == 1)
self.assertTrue(
pc.patterns[hedge('((*/Mn */P.sc) */C (*/M */C))')] == 1)
self.assertFalse(pc.patterns[hedge('(*/Mn */P.sc)')] == 1)
self.assertFalse(pc.patterns[hedge('(*/M */C)')] == 1)
def test_add_with_attributes_search(self):
self.hg.destroy()
edge = hedge('(is graphbrain/1 great/1)')
self.hg.add_with_attributes(edge, {
'p': 1,
'd': 10,
'dd': 20,
'foo': 777,
'bar': -.77,
'xx': 'def'
})
results = set(self.hg.search('(is * *)'))
self.assertEqual(results, set([edge]))
def are_corefs(hg, edge1, edge2, corefs=None):
"""Checks if the two given edges are coreferences."""
if corefs is None:
corefs = {edge1}
for coref_edge in hg.edges_with_edges((hedge(coref_pred), edge1)):
if len(coref_edge) == 3 and coref_edge[0].to_str() == coref_pred:
for item in coref_edge[1:]:
if item not in corefs:
if item == edge2:
return True
corefs.add(item)
if are_corefs(hg, item, edge2, corefs):
return True
return False
def test_primary_2(self):
self.hg.destroy()
edge1 = hedge('((is/M going/P) mary/C (to (the/M gym/C)))')
self.hg.add(edge1, primary=False)
self.assertFalse(self.hg.is_primary(edge1))
self.hg.set_primary(edge1, True)
self.assertTrue(self.hg.is_primary(edge1))
self.assertFalse(self.hg.is_primary(hedge('(is/M going/P)')))
self.hg.set_primary(hedge('(is/M going/P)'), True)
self.assertTrue(self.hg.is_primary(hedge('(is/M going/P)')))
self.assertFalse(self.hg.is_primary(hedge('mary/C')))
self.hg.set_primary(hedge('mary/C'), True)
self.assertTrue(self.hg.is_primary(hedge('mary/C')))
