search

Shreeyak / pytorch-lightning-segmentation-template

Semantic Segmentation on the LaPa dataset using Pytorch Lightning
Apache License 2.0
13 stars 5 forks source link

Use PL's default logging dir structure - need to be logger agnostic #54

Closed Shreeyak closed 3 years ago

Shreeyak commented 3 years ago

currently, the logs are hard-coded to be in project root. we should be able to save at any location (eg - on separate hdd) if given path is relative, make it relative to the project root

Shreeyak commented 3 years ago

Current directory structure is reliant on using wandb to generate a run-id. However, we should be agnostic to logger. What if logger is disabled, or someone is using tensorboard instead?

Hence, we cannot rely on wandb run-id to create dir. Let's use PL's default logging dir structure.

Shreeyak commented 3 years ago

Decided to use wandb to generate the run-id, regardless of the logger selected. This can be improved, by generating the exp dir based on the logger selected.