arya-ebrahimi / options

options

Unofficial Python3 version of "A Laplacian Framework for Option Discovery in Reinforcement Learning" paper.

---

Images from the paper:

Second PVF (left) and its corresponding option (right) in the 4-room domain. Action terminate is depicted in red (top right corner), other actions are depicted as arrows.

Options obtained from the four smallest eigenvectors in the 4-room domain. Action terminate is depicted in red.

Images obtained by running the official Python2 code:

Options obtained from the four smallest eigenvectors in the 4-room domain. Action terminate is depicted using dots.

Value functions in 4-room domain.

PVFs in 4-roon domain.

Images obtained by running the converted Python3 code:

Options obtained from the four smallest eigenvectors in the 4-room domain. Action terminate is depicted using dots.

Value functions in 4-room domain.

PVFs in 4-roon domain.

Q-Learning using options.