Skip to content

yilin-wang/tril

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Trajectory Ranked Imitation Learning

In my senior thesis, I use an inverse reinforcement learning method based on ranked demonstrations for imitation learning, which I call TRIL. This code base is adapted from the original code base for the inverse RL method called T-REX, located here.

Most of my contributions are in the atari folder. The file play_traj.py is used to produce demonstration data, and may be easily modified to produce the desired file names of individual demonstrations. Then, LearnColRewards.py can be used to learn a reward function on the demonstration data, assuming that the demonstrations are ordered in increasing quality, and are generally stored in the learned_models folder.

Use the instructions in the atari folder to run baselines to use regular RL on the learned reward functions. Visualizations of reward functions are in the atari/Visualizations folder, and were produced by visualize_reward.py.

About

Harvard senior thesis, codebase for TRIL (Trajectory Ranked Imitation Learning)

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors