设备不一致问题

[abandol]

您好，MLP还是有变量不在同一设备的问题。

[Dragon1573]

@GDUT-ZJJ 看来这个问题仍然存在，之前在 百度贴吧 测试时都没注意到这里出错了🤣

• 关联 设备不一致 GDUT-ZJJ/NASA#1

---

错误堆栈

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
Cell In[20], [line 11](vscode-notebook-cell:?execution_count=20&line=11)
      [9](vscode-notebook-cell:?execution_count=20&line=9) MAE, RMSE, RE = [], [], []
     [10](vscode-notebook-cell:?execution_count=20&line=10) for seed in range(10):
---> [11](vscode-notebook-cell:?execution_count=20&line=11)     re_list, mae_list, rmse_list, _ = tain(
     [12](vscode-notebook-cell:?execution_count=20&line=12)         LR, feature_size, hidden_size, weight_decay, window_size, EPOCH, seed
     [13](vscode-notebook-cell:?execution_count=20&line=13)     )
     [14](vscode-notebook-cell:?execution_count=20&line=14)     RE.append(np.mean(np.array(re_list)))
     [15](vscode-notebook-cell:?execution_count=20&line=15)     MAE.append(np.mean(np.array(mae_list)))

Cell In[19], [line 38](vscode-notebook-cell:?execution_count=19&line=38)
     [35](vscode-notebook-cell:?execution_count=19&line=35) y = np.reshape(train_y[:, -1] / Rated_Capacity, (-1, 1)).astype(np.float32)
     [37](vscode-notebook-cell:?execution_count=19&line=37) X, y = torch.from_numpy(X).to(device), torch.from_numpy(y).to(device)
---> [38](vscode-notebook-cell:?execution_count=19&line=38) output = model(X)
     [39](vscode-notebook-cell:?execution_count=19&line=39) loss = criterion(output, y)
     [40](vscode-notebook-cell:?execution_count=19&line=40) optimizer.zero_grad()  # clear gradients for this training step

File d:\Repository\NASA\.venv\lib\site-packages\torch\nn\modules\module.py:1736, in Module._wrapped_call_impl(self, *args, **kwargs)
   [1734](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1734)     return self._compiled_call_impl(*args, **kwargs)  # type: ignore[misc]
   [1735](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1735) else:
-> [1736](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1736)     return self._call_impl(*args, **kwargs)

File d:\Repository\NASA\.venv\lib\site-packages\torch\nn\modules\module.py:1747, in Module._call_impl(self, *args, **kwargs)
   [1742](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1742) # If we don't have any hooks, we want to skip the rest of the logic in
   [1743](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1743) # this function, and just call forward.
   [1744](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1744) if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
   [1745](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1745)         or _global_backward_pre_hooks or _global_backward_hooks
   [1746](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1746)         or _global_forward_hooks or _global_forward_pre_hooks):
-> [1747](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1747)     return forward_call(*args, **kwargs)
   [1749](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1749) result = None
   [1750](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1750) called_always_called_hooks = set()

Cell In[18], [line 17](vscode-notebook-cell:?execution_count=18&line=17)
     [15](vscode-notebook-cell:?execution_count=18&line=15) out = self.layer0(x)
     [16](vscode-notebook-cell:?execution_count=18&line=16) for layer in self.layers:
---> [17](vscode-notebook-cell:?execution_count=18&line=17)     out = layer(out)
     [18](vscode-notebook-cell:?execution_count=18&line=18) out = self.linear(out)
     [19](vscode-notebook-cell:?execution_count=18&line=19) return out

File d:\Repository\NASA\.venv\lib\site-packages\torch\nn\modules\module.py:1736, in Module._wrapped_call_impl(self, *args, **kwargs)
   [1734](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1734)     return self._compiled_call_impl(*args, **kwargs)  # type: ignore[misc]
   [1735](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1735) else:
-> [1736](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1736)     return self._call_impl(*args, **kwargs)

File d:\Repository\NASA\.venv\lib\site-packages\torch\nn\modules\module.py:1747, in Module._call_impl(self, *args, **kwargs)
   [1742](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1742) # If we don't have any hooks, we want to skip the rest of the logic in
   [1743](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1743) # this function, and just call forward.
   [1744](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1744) if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
   [1745](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1745)         or _global_backward_pre_hooks or _global_backward_hooks
   [1746](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1746)         or _global_forward_hooks or _global_forward_pre_hooks):
-> [1747](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1747)     return forward_call(*args, **kwargs)
   [1749](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1749) result = None
   [1750](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1750) called_always_called_hooks = set()

File d:\Repository\NASA\.venv\lib\site-packages\torch\nn\modules\container.py:250, in Sequential.forward(self, input)
    [248](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/container.py:248) def forward(self, input):
    [249](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/container.py:249)     for module in self:
--> [250](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/container.py:250)         input = module(input)
    [251](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/container.py:251)     return input

File d:\Repository\NASA\.venv\lib\site-packages\torch\nn\modules\module.py:1736, in Module._wrapped_call_impl(self, *args, **kwargs)
   [1734](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1734)     return self._compiled_call_impl(*args, **kwargs)  # type: ignore[misc]
   [1735](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1735) else:
-> [1736](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1736)     return self._call_impl(*args, **kwargs)

File d:\Repository\NASA\.venv\lib\site-packages\torch\nn\modules\module.py:1747, in Module._call_impl(self, *args, **kwargs)
   [1742](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1742) # If we don't have any hooks, we want to skip the rest of the logic in
   [1743](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1743) # this function, and just call forward.
   [1744](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1744) if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
   [1745](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1745)         or _global_backward_pre_hooks or _global_backward_hooks
   [1746](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1746)         or _global_forward_hooks or _global_forward_pre_hooks):
-> [1747](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1747)     return forward_call(*args, **kwargs)
   [1749](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1749) result = None
   [1750](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/module.py:1750) called_always_called_hooks = set()

File d:\Repository\NASA\.venv\lib\site-packages\torch\nn\modules\linear.py:125, in Linear.forward(self, input)
    [124](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/linear.py:124) def forward(self, input: Tensor) -> Tensor:
--> [125](file:///D:/Repository/NASA/.venv/lib/site-packages/torch/nn/modules/linear.py:125)     return F.linear(input, self.weight, self.bias)

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and cuda:0! (when checking argument for argument mat1 in method wrapper_CUDA_addmm)

运行环境

21:57:18 D:\...\NASA  [main ≡ +5 ~1 -0 !] 0ms pwsh> uv run python -c 'from sys import version; print(version)'
3.9.20 (main, Oct 16 2024, 00:31:33) [MSC v.1929 64 bit (AMD64)]

21:57:24 D:\...\NASA  [main ≡ +5 ~1 -0 !] 51ms pwsh> uv tree -d 1
Resolved 79 packages in 0.67ms
nasa v0.1.0
├── ipykernel v6.29.5
├── matplotlib v3.9.2
├── scikit-learn v1.5.2
├── scipy v1.13.1
├── torch v2.5.1+cu124
├── torchaudio v2.5.1+cu124
├── torchvision v0.20.1+cu124
├── mypy v1.13.0 (group: dev)
└── ruff v0.7.4 (group: dev)

21:58:20 D:\...\NASA  [main ≡ +5 ~1 -0 !] 0ms pwsh> systeminfo.exe

OS Name:                   Microsoft Windows 11 专业工作站版
OS Version:                10.0.22631 N/A Build 22631
OS Manufacturer:           Microsoft Corporation
OS Configuration:          Standalone Workstation
OS Build Type:             Multiprocessor Free
System Manufacturer:       ASUS
System Model:              System Product Name
System Type:               x64-based PC

22:00:19 D:\...\NASA  [main ≡ +5 ~1 -0 !] 0ms pwsh> nvidia-smi.exe
Thu Nov 21 22:00:21 2024       
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 566.03                 Driver Version: 566.03         CUDA Version: 12.7     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                  Driver-Model | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA GeForce RTX 2070      WDDM  |   00000000:01:00.0  On |                  N/A |
|  0%   55C    P0             29W /  131W |    1286MiB /   8192MiB |      1%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

22:00:34 D:\...\NASA  [main ≡ +5 ~1 -0 !] 0ms pwsh> nvcc.exe --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2024 NVIDIA Corporation
Built on Thu_Sep_12_02:55:00_Pacific_Daylight_Time_2024
Cuda compilation tools, release 12.6, V12.6.77
Build cuda_12.6.r12.6/compiler.34841621_0

# CuDNN 使用 v9.5 版本

[Dragon1573]

🎉 解决方案已找到

@abandol @GDUT-ZJJ @XiuzeZhou 👋🏼

Caution

每次运行 MLP.ipynb 前，请 清空所有单元格输出 并 重启内核 ，以此保证您的单元格编号与下方描述一致。

需要修改以下部分：

# 这是 MLP.ipynb 中的第8个 Python 单元格
# 对应编号 In [8]

class Net(nn.Module):
    def __init__(self, feature_size=8, hidden_size=[16, 8]):
        super(Net, self).__init__()
        self.feature_size, self.hidden_size = feature_size, hidden_size
        # 主要改动从下面开始，所有 PyTorch Tensor 或 Module 需要显式移动到目标设备上
        # 否则设备将默认位于 CPU 上，疑似 @XiuzeZhou 未使用 CUDA 设备测试
        self.layer0 = nn.Linear(self.feature_size, self.hidden_size[0]).to(device)
        self.layers = [
            nn.Sequential(
                nn.Linear(self.hidden_size[i], self.hidden_size[i + 1]).to(device),
                nn.ReLU().to(device),
            ).to(device)
            for i in range(len(self.hidden_size) - 1)
        ]
        self.linear = nn.Linear(self.hidden_size[-1], 1).to(device)
        # 核心改动区域结束

    def forward(self, x):
        out = self.layer0(x)
        for layer in self.layers:
            out = layer(out)
        out = self.linear(out)
        return out

# 这是 MLP.ipynb 的第9个 Python 单元格
# 对应编号 In [9]

def tain(
    LR=0.01,
    feature_size=8,
    hidden_size=[16, 8],
    weight_decay=0.0,
    window_size=8,
    EPOCH=1000,
    seed=0,
):
    mae_list, rmse_list, re_list = [], [], []
    result_list = []
    for i in range(4):
        name = Battery_list[i]
        train_x, train_y, train_data, test_data = get_train_test(
            Battery, name, window_size
        )
        train_size = len(train_x)
        print("sample size: {}".format(train_size))

        setup_seed(seed)
        # 核心改动从此处开始
        model = Net(feature_size=feature_size, hidden_size=hidden_size).to(device)

        optimizer = torch.optim.Adam(
            model.parameters(), lr=LR, weight_decay=weight_decay
        )
        criterion = nn.MSELoss().to(device)
        # 核心改动结束

        test_x = train_data.copy()
        loss_list, y_ = [0], []
        for epoch in range(EPOCH):
            X = np.reshape(train_x / Rated_Capacity, (-1, feature_size)).astype(
                np.float32
            )
            y = np.reshape(train_y[:, -1] / Rated_Capacity, (-1, 1)).astype(np.float32)

            X, y = torch.from_numpy(X).to(device), torch.from_numpy(y).to(device)
            output = model(X)
            loss = criterion(output, y)
            optimizer.zero_grad()  # clear gradients for this training step
            loss.backward()  # backpropagation, compute gradients
            optimizer.step()  # apply gradients

            if (epoch + 1) % 100 == 0:
                test_x = train_data.copy()  # 每100次重新预测一次
                point_list = []
                while (len(test_x) - len(train_data)) < len(test_data):
                    x = np.reshape(
                        np.array(test_x[-feature_size:]) / Rated_Capacity,
                        (-1, feature_size),
                    ).astype(np.float32)
                    x = torch.from_numpy(x).to(device)
                    pred = model(
                        x
                    )  # 测试集 模型预测#pred shape为(batch_size=1, feature_size=1)
                    # 核心改动从此处开始
                    # 变量 model 和 x 均位于目标设备（GPU）上，需要将其取回 CPU
                    # 才能对其调用 .data 方法并进一步 .numpy() 转换为 NumPy 数组
                    next_point = pred.cpu().data.numpy()[0, 0] * Rated_Capacity
                    # 核心改动结束
                    test_x.append(next_point)  # 测试值加入原来序列用来继续预测下一个点
                    point_list.append(next_point)  # 保存输出序列最后一个点的预测值
                y_.append(point_list)  # 保存本次预测所有的预测值
                loss_list.append(loss)
                mae, rmse = evaluation(y_test=test_data, y_predict=y_[-1])
                re = relative_error(
                    y_test=test_data, y_predict=y_[-1], threshold=Rated_Capacity * 0.7
                )
                print(
                    "epoch:{:<2d} | loss:{:<6.4f} | MAE:{:<6.4f} | RMSE:{:<6.4f} | RE:{:<6.4f}".format(
                        epoch, loss, mae, rmse, re
                    )
                )
            if (len(loss_list) > 1) and (abs(loss_list[-2] - loss_list[-1]) < 1e-5):
                break

        mae, rmse = evaluation(y_test=test_data, y_predict=y_[-1])
        re = relative_error(
            y_test=test_data, y_predict=y_[-1], threshold=Rated_Capacity * 0.7
        )
        mae_list.append(mae)
        rmse_list.append(rmse)
        re_list.append(re)
        result_list.append(y_[-1])
    return re_list, mae_list, rmse_list, result_list