Validation does not always start at the end of an epoch

[mc4117]

What happened?

If training batches are limited and the total number of batches is not reached then validation does not start. This bug is fixed by setting train batched to null.

What are the steps to reproduce the bug?

Set dataloader.training.limit_batches > number of batches in dataset.

Version

0.3.3

Platform (OS and architecture)

All

Relevant log output

Accompanying data

No response

Organisation

No response

[Rilwan-Adewoyin]

So i think this issue is more related to pytorch lightning not knowing the length of the dataset prior to training run starting.

If it does know the dataset length then regardless of what limit batches value is entered the val loop is not skipped and the end of validation plots are also not skipped

This entails implementing the len property on the Dataset class

[mc4117]

We are now working on implementing a length property on the dataset class