python - Theano: ifelse 类型错误

Question

我正在尝试运行脚本 lstm_ptb.py但它为以下行抛出 TypeError:

shrink_factor = ifelse(T.gt(norm_gparams,max_grad_norm),max_grad_norm/norm_gparams,1.)

这就是这条线试图实现的目标:

 if norm_gparams > max_grad_norm: 
    shrink_factor = max_grad_norm/norm_gparams
 else:
    shrink_factor = 1.

它说:

TypeError: The two branches should have identical types, but they are TensorType(float64, scalar) and TensorType(float32, scalar) respectively. This error could be raised if for example you provided a one element list on the then branch but a tensor on the else branch

请问如何解决这个错误？谢谢

Answer 1

您的问题是由 else 部分的 1. 引起的。默认情况下，它分配为 float32 类型。你只需要转换它:

shrink_factor = ifelse(T.gt(norm_gparams,max_grad_norm),max_grad_norm/norm_gparams,np.float64(1.))

或转换max_grad_norm/norm_gparams值:

shrink_factor = ifelse(T.gt(norm_gparams,max_grad_norm),(max_grad_norm/norm_gparams).astype('float32'),1.)

所以两个值具有相同的类型