piyush2896 / Q-Value-RL

Q-Value (Reinforcement Learning) on Grid World

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Q-Value Iteration - Reinforcement Learning

Q-value is an exact method of solving a Reinforcement Learning problem. The goal of the task is to find expected utility starting in state s, taking action a, and (thereafter) acting optimally.

In mathmetical notations, we calculate this(below equation) for every state-action pair in State-Action space, given an MDP.

Q star of s and a
Src: UC Berkley 2017 Deep RL bootcamp Lecture 1 slides

Task at Hand

The task is to maximize a reward in a world that consists of an agent that can navigate in 4 directions - North, South, East and West. With a 20% of equally likely chance of deviating to left or right from the action asked to perform.

World
Src: UC Berkley 2017 Deep RL bootcamp Lecture 1 slides

Usage

Modify main.json to suit your needs. The key names are self explanatory. Then run python main.py.

You can also create your own <user-defined>.json file with every paramter defined and then run python main.py --json_path <user-defined>.json

About

Q-Value (Reinforcement Learning) on Grid World

License:MIT License


Languages

Language:Python 100.0%