Usage

Create models and datasets

Similar to pytorch lightning, we use LightningModule to implement the model and train, val, and test loop, and use LightningDataModule to implement dataset and dataloaders, for details, see model doc and dataset doc

Config files

Config optimizers and lr schedulers

pytorch lightning does not support multi optimizers and lr schedulers from cli, we add this feature, see doc for detail.

Cross-validation

Set num_folds of the trainer to an int bigger than one to start cross-validation, for details, see doc.

See config file structure, deep update, yaml with merge, and json file for more details.

You can get some examples from project_template

CLI

This project is based on the lightning CLI, so it supports all features from pytorch lightning and lightning CLI, you can get a brief introduction from cli doc.

Wandb Logger

We support logging your code with WandbNamedLogger to control the version of your codes for every experiment. To use Wandb, you have to create an account on their site and login following their doc. We recommend you symlink every file you want to log under the code_dirs folder of your project since WandbNamedLogger has to walk through that folder to find the files you want to log. You can use the following command to do that:

mkdir code_dirs
cd code_dirs
ln -s ../* ../.* .
cd ..

Note that you should run this command in the root folder of your project after you clone this project and you have to symlink any new files and folders you want to log to wandb manually after that. You should not link any files or folders you don’t want to log under the code_dirs folder. If you do not create the code_dirs folder, WandbNamedLogger will walk through the root folder of your project to find the files you want to log, which will be slow, especially if you put your datasets with too many files and folders in your project. Therefore, we recommend you use the above commands to create the code_dirs folder.