torch常规训练方式

通过已有数据的导入进行训练

导入数据与数据增强

import torchvision
train_data=torchvision.dataset.CIFAR10(
	root='./',
	train=True,
	transform=transform_train,
	down=True)

#数据增强
transoform_train=torchvision.transforms.Compose([
	transforms.RandomCrop(32,padding=4),
	transforms.RandomHorizontalFlip(),
	transfoems.RandomRotation(10),
	transforms.Totensor(),
	transforms.Normalize([0.5,0.5,0.5],[0.5,0.5,0.5])])

构建模型

class AutoEncoder(nn.Module):
	def __init__(self):
		super(AutoEncoder,self).__init__()
		self.encoder=nn.Sequential()
		self.decoderr=nn.Sequential()
	def forward(self,x):
		return encode,decode

加载模型和数据

import torch.utils.data as Data
auto=AutoEncoder()

#加载批次数据
train_data_loader=Data.DataLoader(dataset=train_data,batch_size=BATCH_SIZE,shuffle=True)
optimizer=torch.optim.Adam(auto.parameters(),lr=LR)
loss_func=nn.MSELoss()

训练

for epoch in range(EPOCH):
	for i,(x,y) in enumerate(train_data_loade):
		encode,decode=auto(x)
		loss=loss_func(decode,x)
		optimizer.zero_grad()
		loss.backward()
		optimizer.step()
		print('epoch',epoch,'loss',loss)