def inferILPConstrains(self, context, info):
global_key = "global/linguistic/"
root = "sentence"
root_features = ["raw", ]
predictions_on = "word"
prediction_features = ["raw_ready"]
predicates1 = ["FAC", "VEH", "PER", "ORG", "GPE", "LOC", "WEA"]
predicates = []
for item in info:
if item in predicates1:
predicates.append("<"+str(item)+">")
pairs1 = ["ART", "GEN-AFF", "ORG-AFF", "PER-SOC", "METONYMY", "PART-WHOLE", "PHYS"]
pairs = []
for item in info:
if item in pairs1:
pairs.append("<"+str(item)+">")
pairs_on = "pair"
phrase_order1 = ["FAC", "GPE", "PER", "ORG", "LOC", "VEH", "WEA"]
phrase_order = []
for item in info:
if item in phrase_order1:
phrase_order.append(str(item))
sentence = {"words": {}}
with torch.no_grad():
epsilon = 0.00001
for item in predicates:
_list = [_it.cpu().detach().numpy() for _it in context[global_key + predictions_on + "/" + item]]
for _it in range(len(_list)):
if _list[_it][0] > 1-epsilon:
_list[_it][0] = 1-epsilon
elif _list[_it][1] > 1-epsilon:
_list[_it][1] = 1-epsilon
if _list[_it][0] < epsilon:
_list[_it][0] = epsilon
elif _list[_it][1] < epsilon:
_list[_it][1] = epsilon
sentence[item.replace("<", "").replace(">", "")] = _list
sentence['words'][item.replace("<", "").replace(">", "")] = np.log(np.array(_list))
if len(pairs):
sentence['phrase'] = {}
sentence['phrase']['entity'] = {}
sentence['phrase']['raw'] = context[global_key + "phrase/raw"]
sentence['phrase']['tag'] = [_it.item() for _it in context[global_key + "phrase/tag"]]
sentence['phrase']['tag_name'] = [phrase_order[t] for t in sentence['phrase']['tag']]
sentence['phrase']['pair_index'] = context[global_key + "pair/index"]
for item in predicates:
_list = []
for ph in sentence['phrase']['raw']:
_value = [1, 1]
for _range in range(ph[0], ph[1] + 1):
_value[0] = _value[0] * sentence[item.replace("<", "").replace(">", "")][_range][0]
_value[1] = _value[1] * sentence[item.replace("<", "").replace(">", "")][_range][1]
_list.append(np.log(np.array(_value)))
sentence['phrase']['entity'][item.replace("<", "").replace(">", "")] = np.array(_list)
sentence['phrase']['relations'] = {}
for item in pairs:
_list = [np.log(_it.cpu().detach().numpy()) for _it in context[global_key + pairs_on + "/" + item]]
_result = np.zeros((len(sentence['phrase']['raw']), len(sentence['phrase']['raw']), 2))
for _range in range(len(sentence['phrase']['pair_index'])):
indexes = sentence['phrase']['pair_index'][_range]
values = _list[_range]
_result[indexes[0]][indexes[1]][0] = values[0]
_result[indexes[1]][indexes[0]][0] = values[0]
_result[indexes[0]][indexes[1]][1] = values[1]
_result[indexes[1]][indexes[0]][1] = values[1]
for _ii in range(len(sentence['phrase']['raw'])):
_result[_ii][_ii] = np.log(np.array([0.999, 0.001]))
sentence['phrase']['relations'][item.replace("<", "").replace(">", "")] = np.array(_result)
# import pickle
# file = open('data.pkl', 'wb')
#
# # dump information to that file
# pickle.dump(sentence, file)
#
# # close the file
# file.close()
# graph1 = Graph(iri='http://ontology.ihmc.us/ML/ACE.owl', local='./')
# myilpOntSolver = ilpOntSolverFactory.getOntSolverInstance(graph1)
#Using the lConstraints
myilpOntSolver = ilpOntSolverFactory.getOntSolverInstance(app_graph)
if len(pairs):
phrases = [str(_it) for _it in range(len(sentence['phrase']['raw']))]
results = myilpOntSolver.calculateILPSelection(phrases,
sentence['phrase']['entity'],sentence['phrase']['relations'])
else:
tokens = [str(_it) for _it in range(len(sentence['FAC']))]
results = myilpOntSolver.calculateILPSelection(tokens,
sentence['words'])
return self.transform_back(result=results, context=context, helper=sentence)
def __init__(
self,
*args,
**kwargs
):
self.solver = ilpOntSolverFactory.getOntSolverInstance(self, ACELogicalSolver)
def __init__(self, *args, **kwargs):
vocab = None # Vocabulary()
self.model = GraphModel(self, vocab, *args, **kwargs)
self.solver = ilpOntSolverFactory.getOntSolverInstance(
self, AllennlpDataNodeSolver)
self.solver_log_to(None)
def inferILPConstrains(self, context, info):
global_key = "global/linguistic/"
root = "sentence"
root_features = [
"raw",
]
predictions_on = "word"
prediction_features = ["raw_ready"]
predicates1 = [
"ORG", "FAC", "PER", "VEH", "LOC", "WEA", "GPE", "Airport",
"Building-Grounds", "Path", "Plant", "Subarea-Facility",
"Continent", "County-or-District", "GPE-Cluster", "Nation",
"Population-Center", "Special", "State-or-Province", "Address",
"Boundary", "Celestial", "Land-Region-Natural", "Region-General",
"Region-International", "Water-Body", "Commercial", "Educational",
"Entertainment", "Government", "Media", "Medical-Science",
"Non-Governmental", "Religious", "Sports", "Group",
"Indeterminate", "Individual", "Land", "Subarea-Vehicle",
"Underspecified", "Water", "Biological", "Blunt", "Chemical",
"Exploding", "Nuclear", "Projectile", "Sharp", "Shooting",
"WEA-Underspecified"
]
predicates = []
for item in info:
if item in predicates1:
predicates.append("<" + str(item) + ">")
# pairs1 = ["ART", "GEN-AFF", "ORG-AFF", "PER-SOC", "METONYMY", "PART-WHOLE", "PHYS"]
# pairs = []
# for item in info:
# if item in pairs1:
# pairs.append("<"+str(item)+">")
# pairs_on = "pair"
# phrase_order1 = ["ORG", "FAC", "PER", "VEH", "LOC", "WEA", "GPE", "Airport", "Building-Grounds", "Path", "Plant",
# "Subarea-Facility", "Continent", "County-or-District", "GPE-Cluster", "Nation", "Population-Center",
# "Special", "State-or-Province", "Address", "Boundary", "Celestial", "Land-Region-Natural",
# "Region-General", "Region-International", "Water-Body", "Commercial", "Educational", "Entertainment",
# "Government", "Media", "Medical-Science", "Non-Governmental", "Religious", "Sports", "Group",
# "Indeterminate", "Individual", "Land", "Subarea-Vehicle", "Underspecified", "Water", "Biological",
# "Blunt", "Chemical", "Exploding", "Nuclear", "Projectile", "Sharp", "Shooting", "WEA-Underspecified"]
# phrase_order = []
# for item in info:
# if item in phrase_order1:
# phrase_order.append(str(item))
sentence = {"words": {}}
with torch.no_grad():
epsilon = 0.00001
for item in predicates:
_list = [
_it.cpu().numpy()
for _it in context[global_key + predictions_on + "/" +
item]
]
for _it in range(len(_list)):
if _list[_it][0] > 1 - epsilon:
_list[_it][0] = 1 - epsilon
elif _list[_it][1] > 1 - epsilon:
_list[_it][1] = 1 - epsilon
if _list[_it][0] < epsilon:
_list[_it][0] = epsilon
elif _list[_it][1] < epsilon:
_list[_it][1] = epsilon
sentence[item.replace("<", "").replace(">", "")] = _list
sentence['words'][item.replace("<",
"").replace(">", "")] = np.log(
np.array(_list))
# import pickle
# file = open('data.pkl', 'wb')
#
# # dump information to that file
# pickle.dump(sentence, file)
#
# # close the file
# file.close()
# graph1 = Graph(iri='http://ontology.ihmc.us/ML/ACE.owl', local='./')
# myilpOntSolver = ilpOntSolverFactory.getOntSolverInstance(graph1)
#Using the lConstraints
myilpOntSolver = ilpOntSolverFactory.getOntSolverInstance(app_graph)
tokens = [str(_it) for _it in range(len(sentence['FAC']))]
results = myilpOntSolver.calculateILPSelection(tokens,
sentence['words'])
return self.transform_back(result=results,
context=context,
helper=sentence)