Pytorch-lightning: is limit_train_batches shuffle or random

[qmpzzpmq]

hi, I am using limit_train_batches . If it is set, is it means a subdataset of whole train dataset ? similar with torch.utils.data.random_split

[github-actions[bot]]

Hi! thanks for your contribution!, great first issue!

[ydcjeff]

Yes, it is a subset of the train dataset
But, it doesn't similar with random_split

[qmpzzpmq]

@ydcjeff I mean, is it random?

[ydcjeff]

I think it is not random. It is the first limit_train_batches of the train dataset.

[awaelchli]

Yes exactly, @ydcjeff is right. It will fetch batches from the dataloader until it reaches that amount, so your dataset and dataloader settings regarding shuffling will be respected.

[qmpzzpmq]

@awaelchli @ydcjeff thx