PyTorch-Backward

• Backward反向传播

import torch
import numpy as np
import torchvision #torch的视觉包
import torchvision.datasets as datasets
import torchvision.transforms as transforms
from torch.utils.data.dataloader import DataLoader
from torchvision.transforms import ToTensor
import matplotlib.pyplot as plt
import PIL.Image as Image
import os
os.environ["KMP_DUPLICATE_LIB_OK"]="TRUE"
import torch.nn as nn
import torch.nn.functional as F
import torchvision
import torchvision.datasets as datasets
import torchvision.transforms as transforms
class Model(nn.Module):
    def __init__(self):   #初始化
        super(Model, self).__init__()  #调用父类
        self.fc = nn.Linear(4, 3,bias=True)   #线性层
 
    def forward(self, x):
        x = self.fc(x)
        return x
model=Model()
in_features=torch.tensor([1,2,3,4],dtype=torch.float32)
weight_matrix=torch.tensor([
    [1,2,3,4],
    [2,3,4,5],
    [3,4,5,6]
],dtype=torch.float32)
model.fc.weight=nn.Parameter(weight_matrix)
bias_matrix=torch.tensor(
    [1,2,3]
,dtype=torch.float32)
model.fc.bias=nn.Parameter(bias_matrix)
#把weight和bias重新赋值
out=model(in_features)
print(out)
print(model.fc)
print(model.fc.weight)
print(model.fc.bias)
for param in model.parameters():  #输出模型的参数
    print(param)
for name,param in model.named_parameters():
    print(name,':',param.shape)
out=model(in_features).sum() #把结果相加变成标量
print(out)
model.fc.weight.grad=torch.zeros(3,4)
out.backward() #此时就不会报错了
model.fc.weight.grad
in_features=torch.tensor([5,6,7,8],dtype=torch.float32)  #利用新的例子进行测试
out=model(in_features).sum()
print(out)
model.fc.weight.grad=torch.zeros(3,4)
out.backward() #此时就不会报错了
model.fc.weight.grad
in_features=torch.tensor([1,3,5,2],dtype=torch.float32)  #利用新的例子进行测试
out=model(in_features).sum() 
print(out)
model.fc.weight.grad=torch.zeros(3,4)
out.backward() #此时就不会报错了
model.fc.weight.grad

• 利用MNIST数据集做反向传播

import torch
import numpy as np
import torchvision #torch的视觉包
import torchvision.datasets as datasets
import torchvision.transforms as transforms
from torch.utils.data.dataloader import DataLoader
from torchvision.transforms import ToTensor
import matplotlib.pyplot as plt
import PIL.Image as Image
import os
os.environ["KMP_DUPLICATE_LIB_OK"]="TRUE"
import torch.nn as nn
import torch.nn.functional as F
import torchvision
import torchvision.datasets as datasets
import torchvision.transforms as transforms
class CModel(nn.Module):
    def __init__(self):   #初始化
        super(CModel, self).__init__()  #调用父类
        self.fc = nn.Linear(28*28, 10)   #线性层
 
    def forward(self, x):
        x = self.fc(x)
        x=F.softmax(x,dim=1)
        return x
model=CModel()
print(model)
root='D:\Project_Encyclopedia'
mnist=torchvision.datasets.MNIST(root,train=True,transform=ToTensor(),target_transform=None,download=False)
bs=8
mnist_loader=torch.utils.data.DataLoader(dataset=mnist,batch_size=bs,shuffle=True)
batch=next(iter(mnist_loader))
image,labels=batch
grid=torchvision.utils.make_grid(batch[0],nrow=8)#创建一个网络
plt.figure(figsize=(15,15))
plt.imshow(np.transpose(grid,(1,2,0)))
print("labels",batch[1])
model.fc.weight=nn.Parameter(torch.rand(10,784))
input=image.reshape(bs,-1) #转为8×784 这样就可以8×784 与784×10
input.shape
out=model(input)
print(out)
loss=F.cross_entropy(out,label) #交叉熵处理
print(loss)
loss.backward()
print(model.fc.weight.shape)
print(model.fc.weight.grad.shape)
updataweight=model.fc.weight-model.fc.weight.grad #权重系数-梯度
model.fc.weight=nn.Parameter(updataweight)
out=model(input)
model.fc.weight.grad=torch.zeros(10,28*28)
loss=F.cross_entropy(out,label) #交叉熵处理
print(loss)
loss.backward()
updataweight=model.fc.weight-model.fc.weight.grad #权重系数-梯度
model.fc.weight=nn.Parameter(updataweight)
out=model(input)
model.fc.weight.grad=torch.zeros(10,28*28)
loss=F.cross_entropy(out,label) #交叉熵处理
print(loss)