[Edit] Project Title

[Edit] A brief description of what this project does and who it's for.

continualTemplate

This Github repository was set up using the continualTemplate repository as a template. It spins up a PyTorch container (see .devcontainer/Dockerfile), with continualUtils and avalanche packages preinstalled.

Also, this repository uses pluggy hooks to launch training logic in the continualTrain repository.

While a tad complex, setting up the repository like this:

led to a reproducible project with a couple of simple installations
cleanly self-documented how the CL models were trained (just check out training!)
avoided a lot of the redundant code when setting up a new repository.

How Do I Reproduce This Project?

First, in .devcontainer/.env, set the available environment variables. Add in any required packages in local-environment.yml , using conda where possible. This project repository should already come with all the packages needed.

Scripts

devopen.sh

Our workflow uses VSCode and the Dev Containers extension. To use this script, first select the Dev Containers: Install devcontainer CLI command from the Command Palette (F1) to install the devcontainer CLI. Then execute ./devopen.sh path-to-repo-root to set the environment variables and spin up a devcontainer in VSCode.

own.sh

Execute ./own.sh to quickly return ownership to your user (instead of root, thanks to Docker)

Training

Training is configured by a .yml file specifiying the model save path, dataset root path, directory of implementations, and WandB settings. .gitignore has already been configured to ignore training.yml

save_path: "./model_saves/"   # where to save models, will be mounted
dataset_path: "/mnt/"         # where to find datasets, will be mounted
training_dir: "./training/"   # where to find pluggy hook implementations, will be mounted
wandb_enable_logging: true    # enable Weights & Biases logging? 
wandb_api_key: SECRET         # SECRET API KEY

Please ensure that all paths provided in the configuration file are valid directories!

Each file in the training_dir is a python file that implements the following hooks:

get_dataset an Avalanche dataset
get_model a continualUtils inherited torch model
get_criterion a PyTorch criterion
get_strategy an AValanche strategy
get_metadata Must define strategy_name, dataset_name, model_name, wandb_entity, wandb_project_name for logging purposes