CCKS2019_subject_extraction
CCKS2019_subject_extraction copied to clipboard
python3
请问您所用的keras版本是?在自定义loss时总会报错,TypeError: compile() missing 1 required positional argument: 'loss'。 然后,我改成 def myloss(s1_in, ps1): loss1 = K.mean(K.categorical_crossentropy(s1_in, ps1, from_logits=True)) loss2 = K.mean(K.categorical_crossentropy(s2_in, ps2, from_logits=True)) loss = loss1 + loss2 return loss train_model.compile(optimizer=Adam(1e-3),loss=myloss) 跑到代码fit_generator最后报错AttributeError: 'NoneType' object has no attribute 'shape'。
keras版本原因,升到2.2.4就好了,我之前的修改是错的