Dhawgupta/gupta2023behavior

Code for NeurIPS 2023 Spotlight Paper: Behavior Alignment via Reward Function Optimization

Code for Paper Behaviour Alignment Using Learned Reward Functions

The code is prganized into different folders for each experiment. The code for the experiments in the paper can be found in the following folders:

1. barfi-cp : Code for CartPole Experiment
2. barfi-gw : Code for GridWorld Experiment
3. barfi-mc : Code for MountainCar Experiment
4. barfi-mujoco : Code for Mujoco Experiment

The configurations of a experiment is specified using json files, which can be found in the experiments folder of each experiment. To run a configuration(lets for example in GridWorld).

cd barfi-gw
python src/main.py experiments/PaperPlots/good/barfineumann.json 0

To run the 0th configuration in the experiments/PaperPlots/good/barfineumann.json file. The results are stored in the results folder.

python run/pending.py -j experiments/PaperPlots/good/barfineumann.json 

To find all the pending experiments to be run for the barfineumann.json file.

python run/local.py -p src/main.py -j experiments/PaperPlots/good/barfineumann.json -c 8

To run all configurations in the barfineumann.json file using 8 threads in parallel.

After completely running a configuration file, first have to process the data in order to average over all the seeds

python analysis/process_data.py  experiments/PaperPlots/good/barfineumann.json

Then to plot the results

python analysis/learning_curve_plot_final.py  experiments/PaperPlots/good/barfineumann.json

Each folders experiments directory has a folder called PaperPlots which can be used to reproduce the plots in the paper. The good usually refers to good aux reward and bad refers to bad aux reward.