KoyejoLab-ICL-Energy-Based-Models

Setup

(Optional) Update conda:

conda update -n base -c defaults conda

Create a conda environment with the required packages (choosing which based on your CUDA version):

conda env create --file environment_cuda=*.yml -y

To activate the environment:

conda activate icl_ebm

Upgrade pip:

pip install --upgrade pip

Then install any additional packages using pip if you need to:

Then make sure you're logged into wandb:

wandb login

Running

Development & Debugging

The default hyperparameters are set inside src/globals.py. The main entry point is fit_and_score_one.py.

Sweeping

W&B sweeps are included inside sweeps/. To run a sweep, first login to W&B:

wandb login

Then create the sweep:

wandb sweep sweeps/<sweep YAML file>

This will output a sweep ID e.g., ib99560j. Use this sweep ID to run the sweep:

wandb agent <your W&B username>/icl-ebm/<sweep id>

Cluster

Code currently resides on mercury1 and hyperturing1.

Contributing

Code is located inside src/. The main entry point is icl_ebm_train.py. Additional comments:

1. Use black to format your code. See here for more information. To install, pip install black. To format the repo, run black . from the root directory.
2. Use type hints as much as possible.
3. Imports should proceed in two blocks: (1) general python libraries, (2) custom python code. Both blocks should be alphabetically ordered.
4. Plots, when appropriate, should use sns.set_style("whitegrid").

Attribution

Some of this code is based on prior work https://github.com/RylanSchaeffer/KoyejoLab-Nonparametric-Clustering-Associative-Memory-Models