I'm trying to learn more about Alpha Go Zero and came across the book Deep Learning and the Game of Go, accompanied by code in Python here. This repo is my attempt to implement the same in F# while reading the book.
To run the program and watch a game of go beeing played in the terminal:
cd Go
dotnet run
To run the tests:
cd Go.Tests
dotnet test
Currently I have implemented two types of player agents; random playing and playing with Monte Carlo tree search. The follwing table shows results of two agents playing 100 games. The Monte Carlo agent does 100 rollouts before each move, and the agents were playing on a 5x5 board, with komi set to 1.5.
| White player | Black player | White wins | Black wins |
|---|---|---|---|
| Random | Random | 59 | 41 |
| Random | Monte Carlo | 0 | 100 |
| Monte Carlo | Random | 99 | 1 |
| Monte Carlo | Monte Carlo | 52 | 48 |