tf.transpose用pytorch中的函数替换list(tensor)

可以使用torch.stack(x,dim=1)

来达到让list里面的tensor都进行转置并且合并在一块儿

weights = []

for embedding in channel_embeddings:
    weights.append(torch.sum(
        torch.multiply(self.weights['attention'], torch.matmul(embedding, self.weights['attention_mat'])), 1))

score = F.softmax(torch.stack([weights], 1))

有时候要用pytorch重构tensorflow的代码，tf.transpose可以用torch.stack来替换