deep-learning - 大小不匹配，m1：[1 x 5]，m2：[7 x 100] 在 /pytorch/aten/src/TH/generic/THTensorMath.cpp:752

Question

我是 pytorch 的新手并试图创建一个模型，但我收到了这个错误，

AttributeError                            Traceback (most recent call last)
<ipython-input-16-a0f31875b0ba> in <module>()
      1 for t in range(100):
      2     # Forward pass
----> 3     y_pred = model(X_train)
      4 
      5     # Accuracy


AttributeError: 'numpy.ndarray' object has no attribute 'dim'

这是我的代码：

x = np.array([2,4,6,18,20,30,50])
y = x * 2
print(x)
print(y)


shuffle_indices = torch.LongTensor(random.sample(range(0, len(x)), 
len(x)))
x = x[shuffle_indices]
y = y[shuffle_indices]

x = torch.from_numpy(x).float()
y = torch.from_numpy(y.ravel()).long()

# Split datasets
test_start_idx = int(len(x) * 0.75)
X_train = x[:test_start_idx] 
y_train = y[:test_start_idx] 
X_test = x[test_start_idx:] 
y_test = y[test_start_idx:]
print("We have %i train samples and %i test samples." % (len(X_train), len(X_test)))

我们有 5 个训练样本和 2 个测试样本。

class MLP(nn.Module):
    def __init__(self, input_dim, hidden_dim, output_dim):
        super(MLP, self).__init__()
        self.fc1 = nn.Linear(input_dim, hidden_dim)
        self.fc2 = nn.Linear(hidden_dim, output_dim)

    def forward(self, x_in, apply_softmax=False):
        a_1 = F.relu(self.fc1(x_in))
        y_pred = self.fc2(a_1)

        if apply_softmax:
            y_pred = F.softmax(y_pred, dim=1)

        return y_pred

model = MLP(input_dim=len(X_train), 
        hidden_dim=100, 
        output_dim=len(set(y)))
print (model.named_modules)

MLP的绑定方法Module.named_modules（（fc1）：线性（in_features = 7，out_features = 100，bias = True）（fc2）：Linear（in_features = 100，out_features = 7，bias = True））

loss_fn = nn.CrossEntropyLoss()

optimizer = optim.Adam(model.parameters(), lr=1e-3)

def get_accuracy(y_pred, y_target):
    n_correct = torch.eq(y_pred, y_target).sum().item()
    accuracy = n_correct / len(y_pred) * 100
    return accuracy

for t in range(7):
    y_pred = model(X_train)
    _, predictions = y_pred.max(dim=1)
    accuracy = get_accuracy(y_pred=predictions.long(), y_target=y_train)
    loss = loss_fn(y_pred, y_train)
    if t%20==0: 
        print ("epoch: {0} | loss: {1:.4f} | accuracy: {2:.1f}%".format(t, loss, accuracy))
    optimizer.zero_grad()
    loss.backward()
    optimizer.step()

标签： deep-learningpytorchmlp

Answer 1

在您的示例中，X_train是一个numpy.ndarray需要转换为的对象torch.Tensor。因此，您的问题的解决方案是：

y_pred = model(torch.from_numpy(X_train))