batch size 代码

1. 假设已经有了一组numpy 数组；此处框架为PyTorch

x=np.ones((10000,3))
x=torch.Tensor(x).float()
dataset=TensorDataset(x)
dataloader=DataLoader(dataset, batch_size=50,shuffle=True,num_workers=0, drop_last=True)
#。。。
epochs=500
for epoch in range(epochs):
	for batch_idx, data in enumerate (dataloader):
		net.zero_grap()
		data=torch.tensor([item.cpu().detach().numpy() for item in data]).cuda()
        data=data.to(torch.device("cuda"))
		loss=my_criterion(data)
		loss.backward()
		optimizer.step()
		#...
		if batch_idx %100==0:
			torch.save(net.state_dict(),path+“_epoch”+str(epoch)+"myy.pt")