Transferable Neural Network Representations in Reinforcement Learning

To schedule a big experiment on cedar, use slurm.py:

1. Go to your project directory; make sure that it contains the apptainer image pyproject.sif
2. Set your compute requirements in clusters/cedar.json
3. Run the following (after editing it for your job)

module load apptainer 
apptainer exec -C -B .:$HOME pyproject.sif python scripts/slurm.py --cluster clusters/cedar.json --runs 5 -e experiments/Gridworld/E1/P0/DQN-Relu.json 

4. The slurm scripts will be saved in slurm_scripts/. To submit them all, run ./slurm_scripts/submit_all.sh.

---

Acknowledgement: This repository is adapted from erfanMhi/LTA-Representation-Properties, andnp/rl-control-template, and steventango/sparse-feature-transfer.