我想在PyTorch中对张量进行分割。然而，我得到了一个错误信息，因为我无法使分割正常工作。
我希望的行为是将输入数据分割成两个全连接层。然后我想创建一个模型，将这两个全连接层合并成一个。我认为错误是由于x1, x2 = torch.tensor_split(x,2)中的代码错误造成的
import torchfrom torch import nn, optimimport numpy as npfrom matplotlib import pyplot as pltclass Regression(nn.Module): def __init__(self): super().__init__() self.linear1 = nn.Linear(1, 32) self.linear2 = nn.Linear(32, 16) self.linear3 = nn.Linear(16*2, 1) def forward(self, x): x1, x2 = torch.tensor_split(x,2) x1 = nn.functional.relu(self.linear1(x1)) x1 = nn.functional.relu(self.linear2(x1)) x2 = nn.functional.relu(self.linear1(x2)) x2 = nn.functional.relu(self.linear2(x2)) cat_x = torch.cat([x1, x2], dim=1) cat_x = self.linear3(cat_x) return cat_xdef train(model, optimizer, E, iteration, x, y): losses = [] for i in range(iteration): optimizer.zero_grad() # 初始化梯度信息为0 y_pred = model(x) # 进行预测 loss = E(y_pred.reshape(y.shape), y) # 计算损失（调整shape） loss.backward() # 计算梯度 optimizer.step() # 更新梯度 losses.append(loss.item()) # 累积损失值 print('epoch=', i+1, 'loss=', loss) return model, lossesx = np.random.uniform(0, 10, 100) # 随机生成x轴数据y = np.random.uniform(0.9, 1.1, 100) * np.sin(2 * np.pi * 0.1 * x) # 生成正弦波x = torch.from_numpy(x.astype(np.float32)).float() # 将x转换为张量y = torch.from_numpy(y.astype(np.float32)).float() # 将y转换为张量X = torch.stack([torch.ones(100), x], 1) net = Regression()optimizer = optim.RMSprop(net.parameters(), lr=0.01) # 设置优化器为RMSpropE = nn.MSELoss() # 设置损失函数为MSEnet, losses = train(model=net, optimizer=optimizer, E=E, iteration=5000, x=X, y=y)错误信息
/usr/local/lib/python3.7/dist-packages/torch/nn/functional.py in linear(input, weight, bias) 1846 if has_torch_function_variadic(input, weight, bias): 1847 return handle_torch_function(linear, (input, weight, bias), input, weight, bias=bias)-> 1848 return torch._C._nn.linear(input, weight, bias) 1849 1850 RuntimeError: mat1 and mat2 shapes cannot be multiplied (50x2 and 1x32)回答：
Tl;dr
在torch.tensor_split(x,2)中指定dim=1。
解释
x来自于在dim 1上堆叠的两个形状为[100,1]的张量，因此其形状为[100, 2]。应用tensor_split后，你会得到两个形状为[50, 2]的张量。
print(x.shape) # torch.Size([100, 2])print(torch.tensor_split(X,2)[0].shape) # torch.Size([50, 2])错误发生是因为linear1只接受形状为[BATCH_SIZE,1]的张量作为输入，但传递了一个形状为[50, 2]的张量。
如果你的意图是分割随机数数组和全1数组，请将torch.tensor_split(x,2)更改为torch.tensor_split(x,2,dim=1)，这样会生成两个形状为[100,1]的张量。
