[DOC] inconsistent default value for `val_check_interval`

[turbotimon]

📚 Documentation

Inconsistent default value for val_check_interval, where the Core API Documentation claims to be 1.0:

val_check_interval (Union[int, float, None]) – How often to check the validation set. Pass a float in the range [0.0, 1.0] to check after a fraction of the training epoch. Pass an int to check after a fixed number of training batches. An int value can only be higher than the number of training batches when check_val_every_n_epoch=None, which validates after every N training batches across epochs or during iteration-based training. Default: 1.0.

But the Trainer signature says None:

classlightning.pytorch.trainer.trainer.Trainer(*, accelerator='auto', strategy='auto', devices='auto', num_nodes=1, precision=None, logger=None, callbacks=None, fast_dev_run=False, max_epochs=None, min_epochs=None, max_steps=-1, min_steps=None, max_time=None, limit_train_batches=None, limit_val_batches=None, limit_test_batches=None, limit_predict_batches=None, overfit_batches=0.0, val_check_interval=None, check_val_every_n_epoch=1, num_sanity_val_steps=None, log_every_n_steps=None, enable_checkpointing=None, enable_progress_bar=None, enable_model_summary=None, accumulate_grad_batches=1, gradient_clip_val=None, gradient_clip_algorithm=None, deterministic=None, benchmark=None, inference_mode=True, use_distributed_sampler=True, profiler=None, detect_anomaly=False, barebones=False, plugins=None, sync_batchnorm=False, reload_dataloaders_every_n_epochs=0, default_root_dir=None, model_registry=None)

Even tough, further down the same page, it says again Default: 1.0. This seems to be very confusing in situation where I e.g. only want to check every 2 batches by settings check_val_every_n_epoch=2 because the val_check_interval would still be at its default value of 1.0 leading to confusion how often validation happens.

If I understand it correctly, when I explicitly set check_val_every_n_epoch=2 it would take precedence over the val_check_interval=1.0? Therefore, "Default: None" makes more sense (also, it's the Class signature).

Having the default 1.0 in case I set explicit check_val_every_n_epoch=None also makes no sense, since the docs say check_val_every_n_epoch: ... If None, validation will be done solely based on the number of training batches, requiring val_check_interval to be an integer value.