search

andrew-j-levy / Hierarchical-Actor-Critc-HAC-

This repository contains the code to implement the Hierarchical Actor-Critic (HAC) algorithm.
MIT License
253 stars 61 forks source link

Discrete Gridworld #8

Closed ConstantinosM closed 5 years ago

ConstantinosM commented 5 years ago

Hi! Thank you for sharing your work! In the paper you are mentioning that you tested the HAC in a discrete gridworld environment. I assume that you do not need Mujoco for this. Is the code you are sharing here compatible with discrete gridworlds as well?

andrew-j-levy commented 5 years ago

Hello! So this repository implements Hierarchical Actor-Critic (HAC), which is designed specifically for continuous state and action space environments. The other algorithm we introduced, Hierarchical Q-Learning (HierQ), is the discrete state and action space version of HAC and is the algorithm we used for the grid world tasks. I have attached the HierQ code for training agents with 1 (flat), 2, and 3 levels in a four rooms environment. To train a 3-level agent for instance, move to the “HierQ_3_Levels” folder, and enter the command “python3 initialize_HAC --retrain --mix”. After the agent has finished training you can watch the trained agent by entering the command, “python3 initialize_HAC.py --test --show”. At some point I will add this HierQ code to the repo.

HierQ_Code.zip

ConstantinosM commented 5 years ago

Thank you so much for the prompt response and for sharing the code!!!