attributeError：＆＃x27; tuple＆＃x27;对象没有属性＆＃x27; dim＆＃x27;

Question

我正在尝试通过 pytorch库构建变压器网络。我使用的数据集是历史金融市场数据。

 x_train= torch.from_numpy(x_train_tfr)
 x_test= torch.from_numpy(x_test_tfr)
 y_train_tfr = torch.from_numpy(y_train_tfr)
 y_test_tfr = torch.from_numpy(y_test_tfr)

数据准备后，我使用以下代码将X_Train和Y-Train分为12个块：

x_train_split=torch.split(x_train_tfr,12, dim=0)
y_train_split=torch.split(y_train_tfr,12, dim=0)

然后我使用以下代码来训练我的模型：

transformer_model = nn.Transformer(nhead=16, num_encoder_layers=12)
src = x_train_split
tgt = y_train_split
out, state = transformer_model(src, tgt)

但结果如下：

AttributeError                            Traceback (most recent call last)
<ipython-input-64-769f9734fa98> in <module>()
  3 src = x_train_split
  4 tgt = y_train_split
  ----> 5 out, state = transformer_model(src, tgt)

           1 frames
 /usr/local/lib/python3.7/dist-packages/torch/nn/modules/transformer.py in forward(self, src,tgt, src_mask, tgt_mask, memory_mask, src_key_padding_mask, tgt_key_padding_mask,memory_key_padding_mask)
   134         """
   135 
   --> 136         is_batched = src.dim() == 3
   137         if not self.batch_first and src.size(1) != tgt.size(1) and is_batched:
   138             raise RuntimeError("the batch number of src and tgt must be equal")

  AttributeError: 'tuple' object has no attribute 'dim'

如何解决此错误？在我的模型Trainin之前，我必须做任何额外的事情吗？

Answer 1

最佳实践是包括定义x_train_tfr和y_train_tfr的代码。显然，来自X_TRAIN_TFR的一个示例不是您期望的张量，而是元组。数据集实现可能会返回数据本身（张量）以及可能对某些任务有用的其他一些信息（即元数据），并且这些信息作为元组返回。

如果是这种情况，您可以通过执行轻松解决此错误：

src = x_train_split[0]    # or whichever element of the tuple the data tensor is  
tgt = y_train_split 

# Note that model will expect a batch dimension so you'll need to 
# add a batch dimension even if you only have one data example per batch
out, state = transformer_model(src.unsqueeze(0), tgt.unsqueeze(0))