output = Lambda(lambda x: x[:, 1:max_len + 1])(bert_model.output)
model = Model(bert_model.input, output)
def masked_cross_entropy(y_true, y_pred):
"""交叉熵作为loss，并mask掉padding部分的预测
"""
y_true = K.reshape(y_true, [K.shape(y_true)[0], -1])
y_mask = K.cast(K.not_equal(y_true, 0), K.floatx())
cross_entropy = K.sparse_categorical_crossentropy(y_true, y_pred)
cross_entropy = K.sum(cross_entropy * y_mask) / K.sum(y_mask)
return cross_entropy
model.compile(loss=masked_cross_entropy,
optimizer=AdaFactor(learning_rate=1e-3))
model.summary()
def ge_answer(wrong):
"""
解码
:param wrong:
:return:
"""
wrong_token_ids, _ = tokenizer.encode(wrong)
token_ids = wrong_token_ids + [tokenizer._token_mask_id] * max_len + [
tokenizer._token_end_id
]
segemnt_ids = [0] * len(token_ids)
probas = model.predict([np.array([token_ids]),
# 构建模型
model = build_transformer_model(config_path=config_path,
checkpoint_path=checkpoint_path,
application='unilm',
keep_tokens=keep_words)
y_true = model.input[0][:, 1:]
y_mask = model.input[1][:, 1:]
y_pred = model.output[:, :-1]
cross_entropy = K.sparse_categorical_crossentropy(y_true, y_pred)
cross_entropy = K.sum(cross_entropy * y_mask) / K.sum(y_mask)
model.add_loss(cross_entropy)
model.compile(optimizer=AdaFactor(learning_rate=1e-3))
model.summary()
def ge_answer(wrong):
"""
解码
:param wrong:
:return:
"""
wrong_token_ids, _ = tokenizer.encode(wrong)
token_ids = wrong_token_ids + [tokenizer._token_mask_id] * max_len + [
tokenizer._token_end_id
]
segemnt_ids = [0] * len(token_ids)
probas = model.predict([np.array([token_ids]), np.array([segemnt_ids])])[0]