def build(self, input_shape):
super(ConditionalRandomField, self).build(input_shape)
label_size = input_shape[-1]
self._trans = self.add_weight(name='crf_trans',
shape=(label_size, label_size),
initializer='glorot_uniform',
trainable=True)
if self.lr_multiplier != 1:
K.set_value(self._trans, K.eval(self._trans) / self.lr_multiplier)
def on_epoch_end(self, epoch, logs=None):
trans = K.eval(CRF.trans)
wordseg.trans = trans
print(trans)
acc = evaluate(val_data)
if acc > self.best_acc:
self.best_acc = acc
model.save_weights('./best_model.weights')
print('acc is: {:.3f}, best acc is :{:.4f}'.format(acc, self.best_acc))
def build(self, input_shape):
assert len(input_shape) >= 2
input_dim = input_shape[-1]
self._kernel = self.add_weight(shape=(input_dim, self.units),
initializer=self.kernel_initializer,
name='kernel',
regularizer=self.kernel_regularizer,
constraint=self.kernel_constraint)
if self.use_bias:
self._bias = self.add_weight(shape=(self.units,),
initializer=self.bias_initializer,
name='bias',
regularizer=self.bias_regularizer,
constraint=self.bias_constraint)
else:
self.bias = None
self.input_spec = InputSpec(min_ndim=2, axes={-1: input_dim})
self.built = True
if self.lr_multiplier != 1:
K.set_value(self._kernel, K.eval(self._kernel) / self.lr_multiplier)
K.set_value(self._bias, K.eval(self._bias) / self.lr_multiplier)
def on_epoch_end(self, epoch, logs=None):
trans = K.eval(self.model.layers[-1].layers[-1].trans)
NER.trans = trans
print(NER.trans)
f1, precision, recall = evaluate(valid_data, self.model)
# 保存最优
if f1 >= self.best_val_f1:
self.best_val_f1 = f1
self.model.save_weights(self.model_name)
print(
'valid: f1: %.5f, precision: %.5f, recall: %.5f, best f1: %.5f\n'
% (f1, precision, recall, self.best_val_f1))
f1, precision, recall = evaluate(test_data, self.model)
print('test: f1: %.5f, precision: %.5f, recall: %.5f\n' %
(f1, precision, recall))
def on_epoch_end(self, epoch, logs=None):
trans = K.eval(CRF.trans)
NER.trans = trans
print(NER.trans)
f1, precision, recall = evaluate(valid_data)
# 保存最优
if f1 >= self.best_val_f1:
self.best_val_f1 = f1
model.save_weights('./best_model.weights')
print(
'valid: f1: %.5f, precision: %.5f, recall: %.5f, best f1: %.5f\n'
% (f1, precision, recall, self.best_val_f1))
f1, precision, recall = evaluate(test_data)
print('test: f1: %.5f, precision: %.5f, recall: %.5f\n' %
(f1, precision, recall))
mapping = tokenizer.rematch(data, tokens)
token_ids = tokenizer.tokens_to_ids(tokens)
segs = [0] * len(token_ids)
pre = model.predict([[token_ids], [segs]])[0]
labels = self.decode(pre)
words = []
for i, label in enumerate(labels[1:-1]):
if label < 2 or len(words) == 0:
words.append([i + 1])
else:
words[-1].append(i + 1)
return [data[mapping[w[0]][0]:mapping[w[-1]][-1] + 1] for w in words]
wordseg = WordSeg(trans=K.eval(CRF.trans), starts=[0], ends=[0])
def evaluate(data):
"""简单评测"""
total, right = 1e-10, 1e-10
for true in tqdm(data):
pre = wordseg.segment(''.join(true))
w_pre = set(pre)
w_true = set(true)
total += len(w_true)
right += len(w_pre & w_true)
return right / total
if label > 0:
if label % 3 == 1:
starting = True
entities.append([[i], id2label[(label - 1) // 3]])
elif starting:
entities[-1][0].append(i)
else:
starting = False
else:
starting = False
return [(text[mapping[w[0]][0]:mapping[w[-1]][-1] + 1], l)
for w, l in entities]
NER = NamedEntityRecognizer(trans=K.eval(CRF.trans), starts=[0], ends=[0])
def evaluate(data, model):
"""评测函数
"""
X, Y, Z = 1e-10, 1e-10, 1e-10
for d in tqdm(data):
text = ''.join([i[0] for i in d])
R = set(NER.recognize(text, model))
T = set([tuple(i) for i in d if i[1] != 'O'])
X += len(R & T)
Y += len(R)
Z += len(T)
f1, precision, recall = 2 * X / (Y + Z), X / Y, X / Z
return f1, precision, recall
