CS394R_Final

Chloe Chen and Logan Persyn

Final Report: final_report.pdf

Video: YouTube

Environment

1. Create new conda env with python 3.11: conda create --name env_name python=3.11
2. Install all dependencies with pip: pip install -r requirements.txt or pip install -r tacc_requirements.txt
3. Use pip list --format=freeze > requirements.txt to create environment.

Training

• The respective tacc_train_* file has the commands and arguments to train agents.

Testing

• Different noise weight tests can be run with noise_weight_test_with_*.sh files.

Watching

• An example command to watch the noisy-trained SAC, watch mode is controlled by --watch and --render fps flags, both must be present: python .\tianshou\atari_sac.py --watch --resume-path .\final_policies\sac_trained_on_noisy_env.pth --logdir ./temp --render 0.01

RAM

• All code and results from RAM model training are in the ram folder. To train the model, use python sac_ram.py --train. 500 episodes should take around 2-4 hours to run depending on your computation power.
• python sac_ram.py --test can be used to test already trained model.