How to reload data loader inside an epoch.

[jfkback]

I am using PL with a PyTorch data loader and am passing it to the trainer. I would like to have the data loader update during training so that new data is used. I would just use the reload_dataloaders_every_epoch option, but I want this to happen in the middle of an epoch as one epoch is very long and I want the data to be updated faster than it takes for an epoch to complete. Is there any way to do this? I can understand it might be difficult as the data loader will be iterated over while training so changing it could lead to problems. One other possibility would be to shorten the epochs so that each epoch only does a small part of the total data, is there a way to partition a data loader like this?

[Borda]

I think that you can bypass it by adding this dataloader reload in the validation end and set Trainer's argument val_check_interval to perform the validation twice per epoch