Why is the default implementation for train_dataloader in DataHooks a warning? #8734
-
|
The default implementation for these is logging a warning that nothing is implemented. https://github.com/PyTorchLightning/pytorch-lightning/blob/963c26764682fa4cf64c93c5a7572ae0040e9c32/pytorch_lightning/core/hooks.py#L529 Why isn’t the default implementation to raise a NotImplementedError? This would make errors much clearer in case users forget to override these hooks |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
|
I believe this is just legacy code. No real reason. It's in the original implementation (to bolts!) Lightning-Universe/lightning-bolts@797464c Feel free to try changing it :) |
Beta Was this translation helpful? Give feedback.
-
|
filed #8752 as a followup |
Beta Was this translation helpful? Give feedback.
I believe this is just legacy code. No real reason. It's in the original implementation (to bolts!)
Lightning-Universe/lightning-bolts@797464c
Feel free to try changing it :)