/model-based-RL

Primary LanguagePythonMIT LicenseMIT

Model-Based-RL

AlphaZero

Alphagif

This is the final trained Network in play against a Human Opponent.

Play

To play against the Algorithm, clone this repository on your local machine, make sure you have python 3 or higher as well as PyTorch installed.

Then Run the Python Script Play.py in the AlphaZero Folder

Overview

AlphaZero uses a Neural Network as a value approximator, this Network works in Conjunction with Monte Carlo Tree Search (MCTS)

Network Architecture

ArchiDiag

The Exact architecture can be found in Nets.py

The Graph Below shows how the Networks Perform Compared to their previous networks, during training:

HistGraph

This graph implies that, except for network number 5 (compared to network number 4), there was an improvement every network.