class UnitParser:
def __init__(self, parser_model_file = None):
if parser_model_file:
self.model = ParserModel(parser_model_file)
else:
self.model = ParserModel()
def __get_scores_for_MST(self, sent, unit, model, map_func, feats):
scores = {}
unigrams = make_unigram_features(sent)
for d in unit:
for h in unit + [0]:
if h != d:
vector = make_features_for_parser(sent, unigrams, h, d, map_func, feats)
scores[(h,d)] = (model.score(vector), [(h,d)])
return scores
def train(self, instances, model_file, epochs = 10):
self.model.make_weights()
print 'start training ...'
q = 0
for epoch in xrange(epochs):
correct = 0
total = 0
for (gold, head_vectors) in iter(instances):
q += 1
pred = self.model.predict(head_vectors)
if gold != pred:
self.model.update(head_vectors[gold], head_vectors[pred], q)
else:
correct += 1
total += 1
print '\nepoch %d done, %6.2f%% correct' % (epoch,100.0*correct/total)
self.model.average(q)
self.model.save(model_file)
# def train_CLE(self, instances, model_file, epoch = 10):
# self.model.make_weights()
# for epoch in xrange(epochs):
# correct = 0
# total = 0
# for sent in instances:
def train_CLE(self, instances, model_file, epochs = 10):
print 'start training ...'
self.model.make_weights()
q = 0
for epoch in xrange(epochs):
correct = 0
total = 0
for (gold_arcs, vectors) in iter(instances):
q += 1
scores = {}
for a in vectors:
scores[a] = (self.model.score(vectors[a]), [a])
pred_arcs = MST(scores).edges()
gold_heads = dict([(d, h) for (h, d) in gold_arcs])
pred_heads = dict([(d, h) for (h, d) in pred_arcs])
for d in gold_heads:
if gold_heads[d] != pred_heads[d]:
self.model.update(vectors[(gold_heads[d], d)], vectors[(pred_heads[d], d)], q)
else:
correct += 1
total += 1
print '\nepoch %d done, %6.2f%% correct' % (epoch,100.0*correct/total)
self.model.average(q)
self.model.save(model_file)
def predict(self, sent, unit, feats):
score = self.__get_scores_for_MST(sent, unit, self.model, self.model.map_feature, feats)
graph = MST(score)
sent.add_unitheads(graph.edges())
# use to check unit accuracy
# sent.add_heads(graph.edges())
return sent
