注； 在调试时发生了变更的_tensor.py的678行中添加了detech
importtorchfromtorch.utilsimportdata #获取迭代数据from torch.autogradimportvariable # importtorchvisionfromtorchvision.datasets变量检索数据集import matplotlib.pyplot as plt# 数据集预处理data _ TF=torch vision.transforms.com pose ([ torch vision.transforms.to tensor ] torch vision.transforms.) 0.5 ) ] )数据路径=r ' d : (机械学习)数据' #获取数据集traiing transform=data_tf，download=true (测试Transform=data_tath，train=False，transform=data _ TF download=true (train _ loader=data.data loader ) train shuffle=True ) test_loader shuffle=true(#网络结构classCNNnet(torch.nn.module ) :def_init_ ) self ) 33365365306; self(_init_ ) self.conV1=torch.nn.sequential ) torch.nn.conv2d(in_channels=1，out_channels ) ) torch.nn.batchnorm2d(16 (，torch.nn.ReLU ) ) self.conV2=torch.nn.sequential ) torch.nn.conv2d ) 16， torch.nn.ReLU () ) self.conV3=Torch.nn.sequential ) torch.nn.conv 2d (32，64，3，2，1 )， Torch.nn.Baaar torch.nn.ReLU () ) self.conV4=Torch.nn.sequential ) torch.nn.conv 2d (64，64，2，2，2，0 ) Torch.nn.Baaar torch.nn.ReLU () ) self.ml P1=torch.nn.linear (2*2* 64，100 ) self.MLP2=torch.nn.Linn 10 ) ) ) 652 x65:x=self.conV1(x ) x=self.conv2(x ) x=self.conv3(x ) x=self.conv4(x ) print ) model ) loss _ func=torch.nn.crossentropyloss ) ) opt=lr=0.001 ) loss_count=[]forepochinrange(2) y )在枚举(train _ loader ) 3360batch_x=28 ) batch_y=variable(y ) #torch.size ) [128] )最终输出out=model 10]使用检索的batch_y ) #优化器反向传播损耗opt.zero_grad(#上次剩余更新参数值loss.backward ) #误差，计算参数更新值opt.step(# 将参数更新值应用于net的parmeters上的ifI==0:loss_count.append ) loss ) print({}:(t'.format ) I}的r ' d : CNN ' ) if i % 100==0: for a， bintest _ loader 3360 test _ x=variable (a ) test_y=Variable(b ) b ) out=model(test_x ) #print ) test _ ou ot test_y ) accuracy=torch.max(out，1 ) ).numpy ) test_y accuracy.mean () ) break PLT.figure (py torch _ CNN