Hippo模型预测外汇价格时的张量维度与训练报错问题

背景 #

我正在用Hippo模型做外汇价格预测，数据集包含datetime列和开盘价列。

遇到的问题 #

• 更新状态u的第二维度应为batch size，预期形状为[64,1]，我已经将其设置为1，但实际输出形状仍为[1,64]；
• 训练阶段出现RuntimeError：pred张量尺寸（64）与y_train张量尺寸（4322）在非单例维度0上不匹配。

错误详情 #

Cell In[18], line 17, in train(X_train, y_train, model, loss_fn, optimizer, batch_size, device)
     14 print('pred.shpae:',pred.shape)
     15 print('y_train',y_train.shape)
---> 17 loss = loss_fn(pred, torch.tensor(y_train))
     18 optimizer.zero_grad()
     19 loss_backward()

File ~\anaconda3\lib\site-packages\torch\nn\modules\module.py:1511, in Module._wrapped_call_impl(self, *args, **kwargs)
   1509     return self._compiled_call_impl(*args, **kwargs)  # type: ignore[misc]
   1510 else:
-> 1511     return self._call_impl(*args, **kwargs)

File ~\anaconda3\lib\site-packages\torch\nn\modules\module.py:1520, in Module._call_impl(self, *args, **kwargs)
   1515 # If we don't have any hooks, we want to skip the rest of the logic in
   1516 # this function, and just call forward.
   1517 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
   1518         or _global_backward_pre_hooks or _global_backward_hooks
   1519         or _global_forward_hooks or _global_forward_pre_hooks):
-> 1520     return forward_call(*args, **kwargs)
   1522 try:
   1523     result = None

File ~\anaconda3\lib\site-packages\torch\nn\modules\loss.py:535, in MSELoss.forward(self, input, target)
    534 def forward(self, input: Tensor, target: Tensor) -> Tensor:
--> 535     return F.mse_loss(input, target, reduction=self.reduction)

File ~\anaconda3\lib\site-packages\torch\nn\functional.py:3338, in mse_loss(input, target, size_average, reduce, reduction)
   3335 if size_average is not None or reduce is not None:
   3336     reduction = _Reduction.legacy_get_string(size_average, reduce)
-> 3338 expanded_input, expanded_target = torch.broadcast_tensors(input, target)
   3339 return torch._C._nn.mse_loss(expanded_input, expanded_target, _Reduction.get_enum(reduction))

File ~\anaconda3\lib\site-packages\torch\functional.py:76, in broadcast_tensors(*tensors)
     74 if has_torch_function(tensors):
     75     return handle_torch_function(broadcast_tensors, tensors, *tensors)
---> 76 return _VF.broadcast_tensors(tensors)

RuntimeError: The size of tensor a (64) must match the size of tensor b (4322) at non-singleton dimension 0

报错的train方法代码 #

def train(X_train, y_train, model, loss_fn, optimizer, batch_size = None, device = None):
    size = len(X_train)
    if device is None:
        device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

    model.to(device)
    if batch_size is None:
#         X_train_tensor= torch.tensor(X_train, dtype = torch.float32).to(device)
#         y_train_tensor = torch.tensor(y_train, dtype = torch.float32).to(device)
    

        model.train()
        pred = model(X_train)[0]  
        print('pred.shpae:',pred.shape)
        print('y_train',y_train.shape)
        
        loss = loss_fn(pred, torch.tensor(y_train))
        optimizer.zero_grad()
        loss_backward()
        optimizer.step()
        loss_value = loss.item()
        print(f'loss:{loss_value:>7f}, [{size:>5d}/{size:>5d}]')
    else:
        dataset = TimeSeriesDataset(X_train, y_train)
        dataloader = DataLoader(dataset, batch_size = batch_size, shuffle = False)
        
        model.train()
       
        for batch_idx, (X_batch, y_batch )in enumerate(dataloader):
            X_batch, y_batch = X_batch.to(device,dtype = torch.float32), y_batch.to(device,dtype = torch.float32)
        
        # computing prediction error
        pred = model(X_batch)
        loss_ = loss_fn(pred, y_batch)

        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        if batch%10==0:
            loss_value = loss.item()
            current = batch_idx * len(X_batch)
            print(f"loss: {loss_value:>7f}  [{current:>5d}/{len(dataset):>5d}]")

内容的提问来源于stack exchange，提问作者yazi