TypeError：mul() 参数“其他”（位置 1）必须是张量，而不是 ReLU

Question

我想在 fc1 和 fc2 层之间添加一个torch.nn.ReLU() 层。

原码：

型号：

# ...
self.fc1 = nn.Linear(4096, 256)
self.fc2 = nn.Linear(256, 4096)
# ...

def forward(...):
    # ...
    x = x.view(-1, 4096)
    x = self.fc1(x))
    if a7 is not None:
        x = x * a7.squeeze()
    # ...

我试过了

# ...
x = x.view(-1, 4096)
x = nn.ReLU(self.fc1(x)))
if a7 is not None:
    x = x * a7.squeeze()
# ...

然后弹出这个错误。

Answer 1

我的回答假设__init__ 是一个错字，应该是forward。如果不是这样，请告诉我，我会删除它。

import torch
from torch import nn

class SimpleModel(nn.Module):
  def __init__(self, with_relu=False):
    super(SimpleModel, self).__init__()
    self.fc1 = nn.Sequential(nn.Linear(3, 10), nn.ReLU(inplace=True)) if with_relu else nn.Linear(3, 10)
    self.fc2 = nn.Linear(10, 3)

  def forward(self, x):
    x = self.fc1(x)
    print(torch.min(x))  # just to show you ReLU is working...
    return self.fc2(x)

# Model without ReLU
net_without_relu = SimpleModel(with_relu=False)
print(net_without_relu)

# Model with ReLU
net_with_relu = SimpleModel(with_relu=True)
print(net_with_relu)

# random input data
x = torch.randn((5, 3))
print(x)

# we expect it to print something < 0
output1 = net_without_relu(x)

# we expect it to print 0.
output2 = net_with_relu(x)

您可以查看 Colab 上运行的以下代码：https://colab.research.google.com/drive/1W3Dh4_KPd3iABx5FSzZm3tilm6tnJh0v

---

按照您的尝试使用：

x = nn.ReLU(self.fc1(x)))

您可以使用函数式 API：

from torch.nn import functional as F

# ...
x = F.relu(self.fc1(x)))

Answer 2

您不应该在__init__ 中执行任何查看方法。 Init 应该保持你的结构。 例如，这是从 AlexNet __init__复制而来的

nn.Linear(4096, 4096),
nn.ReLU(inplace=True),
nn.Linear(4096, num_classes),

但是，您的转发方法可能包含重塑、计算、函数。

---

nn.Sequential 应该是 __init__ 的一部分，就像在 AlexNet 中一样：

class AlexNet(nn.Module):

    def __init__(self, num_classes=1000):
        super(AlexNet, self).__init__()
        self.features = nn.Sequential(
            nn.Conv2d(3, 64, kernel_size=11, stride=4, padding=2),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=3, stride=2),
            nn.Conv2d(64, 192, kernel_size=5, padding=2),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=3, stride=2),
            nn.Conv2d(192, 384, kernel_size=3, padding=1),
            nn.ReLU(inplace=True),
            nn.Conv2d(384, 256, kernel_size=3, padding=1),
            nn.ReLU(inplace=True),
            nn.Conv2d(256, 256, kernel_size=3, padding=1),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=3, stride=2),
        )
        self.classifier = nn.Sequential(
            nn.Dropout(),
            nn.Linear(256 * 6 * 6, 4096),
            nn.ReLU(inplace=True),
            nn.Dropout(),
            nn.Linear(4096, 4096),
            nn.ReLU(inplace=True),
            nn.Linear(4096, num_classes),
        )

    def forward(self, x):
        x = self.features(x)
        x = x.view(x.size(0), 256 * 6 * 6)
        x = self.classifier(x)
        return x

然后你可以使用类属性self.features，self.classifier 转发。

_{注意：这是来自 PyTorch 0.4 的 AlexNet 的旧模型，但相当简单，逻辑相同}