Question

我正在尝试在conv1d中创建自定义权重，如下所示：

import torch
from torch import nn
conv = nn.Conv1d(1,1,kernel_size=2)
K = torch.Tensor([[[0.5, 0.5]]])
with torch.no_grad():
conv.weight = K

但是我得到了错误

"File “D:\ProgramData\Miniconda3\envs\pytorchcuda102\lib\site-packages\torch\nn\modules\module.py”, line 611, in setattr
raise TypeError(“cannot assign ‘{}’ as parameter ‘{}’ "
TypeError: cannot assign ‘torch.FloatTensor’ as parameter ‘weight’ (torch.nn.Parameter or None expected)”

我在做什么错了？

Answer 1

你很近。请注意，您不需要调用“ with torch.no_grad（）”，因为在权重分配过程中不会计算梯度。您需要做的就是将其删除，然后调用“ conv.weight.data”而不是“ conv.weight”，以便您可以访问基础参数值。请参见下面的固定代码：

import torch
from torch import nn
conv = nn.Conv1d(1,1,kernel_size=2)
K = torch.Tensor([[[0.5, 0.5]]])
conv.weight.data = K

Answer 2

根据讨论here，将代码更新为包含torch.nn.Parameter()，这基本上使得权重可以作为优化器中的参数来识别。

import torch
from torch import nn
conv = nn.Conv1d(1,1,kernel_size=2)
K = torch.tensor([[[0.5, 0.5]]]) #use one dimensional as per your conv layer
conv.weight = nn.Parameter(K) #use nn.parameters

对于更大和更复杂的模型，您可以看到这个使用pytorch state_dict()方法的玩具示例。

import torch
import torch.nn as nn
import torchvision

net = torchvision.models.resnet18(pretrained=True)

pretrained_dict = net.state_dict()
conv_weights = pretrained_dict['conv1.weight'] #64,3,7,7

new = torch.tensor((), dtype=torch.int32)
new = new.new_ones(conv_weights.shape) #assigning all ones

pretrained_dict['conv1.weight'] = new


net.load_state_dict(pretrained_dict)
param = list(net.parameters())

print(param[0])

conv1d pytorch中的自定义重量初始化

2 个答案: