nn.Conv2d(3,6,kernel_size 5,stride 1,padding 0),
nn.AvgPool2d(kernel_size 2,stride 2,padding 0), #长和宽各变为原来的1半
nn.Conv2d(6,16,kernel_size 5,stride 1,padding 0),
nn.AvgPool2d(kernel_size 2,stride 2,padding 0)
# flatten 在forward里实现
#fc unit
self.fc_unit nn.Sequential(
nn.Linear(16*5*5,120),
nn.ReLU(),
nn.Linear(120,84),
nn.ReLU(),
nn.Linear(84,10)
#[B,16,5,5] 打平
tmp torch.randn(2,3,32,32)
out self.conv_unit(tmp)
print(out.shape)
#交叉熵 分类问题用交叉熵 逼近regresssion用mse
self.criteon nn.CrossEntropyLoss()
#pytorch自动根据你写的forward 自己弄Backward
def forward(self,x):
batchsz x.size(0) #x.size返回一个List [b,3,32,32]
# [b,3,32,32] - [b,16,5,5]
x self.conv_unit(x)
# [b,16,5,5] - [b,16*5*5]
x x.view(batchsz,16*5*5) #设置好了第一维 其它内容自动归到第二个维度
# [b,16*5*5] - [b,10]
logits self.fc_unit(x)
#[b,10]
#pred F.softmax(logits,dim 1)
#.CrossEntropy 包含softmax 和 error
#F是函数不需要初始化 nn需要初始化 且nn内部定义了__call__ 可以自动调用forward
return logits
if __name__ __main__ :
#测试代码
net Lenet5()
tmp torch.randn(2,3,32,32)
out net(tmp)
print( lenet out:{} .format(out.shape))
import torch.optim as optim
from torchvision import datasets,transforms
from torch.utils.data import random_split,DataLoader
import torch.nn as nn
import torch.nn.functional as F
import torch
from pytorch_39_Lenet import Lenet5 #导入lenet
batch_size1 32
def main():
#一次加载一张
cifar_train datasets.CIFAR10( cifar ,True,transform transforms.Compose([
transforms.Resize((32,32)),
transforms.ToTensor()
]),download True)
#多线程 批量读取
cifar_train DataLoader(cifar_train,batch_size batch_size1,shuffle True)
cifar_test datasets.CIFAR10( cifar ,True,transform transforms.Compose([
transforms.Resize((32,32)),
transforms.ToTensor()
]),download True)
#多线程 批量读取
cifar_test DataLoader(cifar_test,batch_size batch_size1,shuffle True)
x,label iter(cifar_train).next()
print( x: ,x.shape, label: ,label.shape)
device torch.device( cuda )
model Lenet5().to(device)
criteon nn.CrossEntropyLoss().to(device)
optimizer optim.Adam(model.parameters(),lr 1e-3)
print(model)
model.train()
for epoch in range(1000):
for batchidx,(x,label) in enumerate(cifar_train):
#[b,3,32,32]
#[b]
x,label x.to(device),label.to(device)
logits model(x)
#logits:[b,10]
#label: [b]
loss criteon(logits,label)
#backprop
optimizer.zero_grad() #不清零就是累加
loss.backward()
optimizer.step()
print(epoch,loss.item())
# test
无论requires_grad是Ture还是False 反向传播时都不会自动求导。不会自动构建图
test可以实现一定速度的提升 并节省一半的显存 因为其不需要保
存梯度。
因为会有一些dropout等操作 所以在训练时候需要设置model.train()
测试的时候只要设置model.eval()
model.eval()
with torch.no_grad():
total_correct 0
total_num 0
for x, label in cifar_test:
#[b,3,32,32]
#[b]
x,label x.to(0), label.to(device)
#[b,10]
logits model(x)
#[b] vs [b] - scalar tensor
pred logits.argmax(dim 1)
total_correct torch.eq(pred,label).float().sum()
total_num x.size(0)
acc total_correct / total_num
print(epoch,acc)
main()
