a question about train_step

[minhoooo1]

I was reading other people's code and found PL, an excellent framework. But I have one doubt about the training step .Let me show you the code related to dataloader

When I use 1000 training pictures, In the first figure len(self.files) = 1000 ,len(AEI_Dataset) = 5000 and batch_size = 20 ，train_ step should be 250, but it is 125000250 .You can see the figure below 。

I hope a kind person can help me answer it. Thank you very much

[rohitgr7]

hey @minhoooo1 !

you can multiplied the total len by 5 inside your dataset's __len__ method.
Also, the main progress bar shows the total steps (including both train & val) in the main progress bar. You can verify the total count by printing len(dataloader) to check the length.

[minhoooo1]

I tried, but the number of steps on the main process is different from len(dataloader)。I still don't know total steps calculation method at present. Do you have any documents? If one sentence is clear, it's best。Thank you！

…

------------------ 原始邮件 ------------------ 发件人: "PyTorchLightning/pytorch-lightning" ***@***.***>; 发送时间: 2022年2月21日(星期一) 下午4:20 ***@***.***>; ***@***.******@***.***>; 主题: Re: [PyTorchLightning/pytorch-lightning] a question about train_step (Discussion #11989) hey @minhoooo1 ! you can multiplied the total len by 5 inside your dataset's __len__ method. Also, the main progress bar shows the total steps (including both train & val) in the main progress bar. You can verify the total count by printing len(dataloader) to check the length. — Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android. You are receiving this because you were mentioned.Message ID: ***@***.***>