Pytorch使用过程错误与解决 -汇总~

error1：关键词 copy tensor

报错信息：

UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).

解决办法1：
*** 当转换某个变量为tensor时，尽量使用torch.as_tensor()***

原错误代码：
state_tensor_list = [torch.tensor(i) for i in batch.state]

修改为：
state_tensor_list = [torch.as_tensor(i) for i in batch.state]

解决办法2：
*** 当转换某个变量x为tensor时，尽量使用x.clone().detach() or x.clone().detach().requires_grad_(True) ***

原错误代码：
state_tensor_list = [torch.tensor(i) for i in batch.state]

修改为：
state_tensor_list = [i.clone().detach() for i in batch.state]

error2：关键词 张量相加

参考链接:

报错信息：
RuntimeError: The size of tensor a (3) must match the size of tensor b (4) at non-singleton dimension 1
解决办法：
检查张量维度！！！

错误代码
next_state_values = torch.tensor([0.7056, 0.7165, 0.6326])
state_action_values=torch.tensor([[ 0.1139,  0.1139,  0.1139,  0.1139],
        [ 0.0884,  0.0884,  0.0884,  0.0884],
        [ 0.0019,  0.0019,  0.0019,  0.0019]])
print(next_state_values.shape)
print(state_action_values.shape)
print(next_state_values.size())
print(state_action_values.size())
next_state_values + state_action_values

结果：
torch.Size([3])
torch.Size([3, 4])
torch.Size([3])
torch.Size([3, 4])

修改代码：
next_state_values = torch.tensor([0.7056, 0.7165, 0.6326])
state_action_values=torch.tensor([[ 0.1139,  0.1139,  0.1139,  0.1139],
        [ 0.0884,  0.0884,  0.0884,  0.0884],
        [ 0.0019,  0.0019,  0.0019,  0.0019]]).max(1)[0]
print(next_state_values.shape)
print(state_action_values.shape)
print(next_state_values.size())
print(state_action_values.size())
next_state_values + state_action_values

结果：
torch.Size([3])
torch.Size([3])
torch.Size([3])
torch.Size([3])
tensor([0.8195, 0.8049, 0.6345])

error3：关键词 nn.Linear()的使用

参考链接:

报错1：

报错：RuntimeError: expected scalar type Long but found Float

报错代码：

import torch
import torch.nn as nn
class Net(nn.Module):
    def __init__(self,state_dim,mid_dim,action_dim):
        super().__init__()
        self.net = nn.Sequential(nn.Linear(state_dim, mid_dim))
    def forward(self,state):
        res = self.net(state)
        return res
net = Net(9,5,4)
print(net)
current_state = torch.tensor([0,0,0,0,0,0,0,0,0])
print(current_state.shape)
action = net(current_state)
print(action)

结果：
Net(
  (net): Sequential(
    (0): Linear(in_features=9, out_features=5, bias=True)
  )
)
torch.Size([9])
torch.Size([9])

错误原因：

将一维张量转换成二维张量后才能输入

报错2：

报错：RuntimeError: expected scalar type Float but found Long

报错代码：

import torch
import torch.nn as nn
class Net(nn.Module):
    def __init__(self,state_dim,mid_dim,action_dim):
        super().__init__()
        self.net = nn.Sequential(nn.Linear(state_dim, mid_dim))
    def forward(self,state):
        res = self.net(state)
        return res
net = Net(9,5,4)
print(net)
current_state = torch.tensor([0,0,0,0,0,0,0,0,0])
print(current_state.shape)
current_state = current_state.view(1,9)
print(current_state.shape)
action = net(current_state)
print(action)
结果：
Net(
  (net): S
  equential(
    (0): Linear(in_features=9, out_features=5, bias=True)
  )
)
torch.Size([9])
torch.Size([1, 9])

错误原因：

传入的张量数据类型应该为float32

解决办法

错误原因：

1、将张量转为二维使用.view(1,shape)
2、指定张量数据类型 dtype=torch.float32

正确代码

import torch
import torch.nn as nn
class Net(nn.Module):
    def __init__(self,state_dim,mid_dim,action_dim):
        super().__init__()
        self.net = nn.Sequential(nn.Linear(state_dim, mid_dim))
    def forward(self,state):
        res = self.net(state)
        return res
net = Net(9,5,4)
print(net)
current_state = torch.tensor([0,0,0,0,0,0,0,0,0],**dtype=torch.float32**)
print(current_state.shape)
**current_state = current_state.view(1,9)**
print(current_state.shape)
action = net(current_state)
print(action)