Book about Neural Networks for Chess

[reflectionofpower]

Looks to be an interesting read. Thank You

[Fabio Gobbato]

Thank you, it's a very nice work!

[Mike Sherwin]

Slightly about a year ago I got interested in how AlphaGo, AlphaGo Zero, AlphaZero and then later Stockfish NNUE actually worked. I've looked around but found the papers not easy to read (they take ideas from various domains), and started to take notes.

In the end I wrote a book and decided to give it away as a free PDF. So if you are interested in the topic, I invite you to have a look at

https://github.com/asdfjkl/neural_network_chess

Hi Dominik,

I am very excited about your book! Maybe now I'll be able to write my own NN evaluator for for my new engine Bricabrac. However, I feel I have to beat my own drum just a bit. My engine RomiChess has had deep RL from January of 2006. At Leo Dijksman's (RIP) WBEC RomiChess climbed from class D to class B and was about to promote to class A when Leo suffered a hard drive failure and lost Romi's learn file.

Here is a quote from the CPW.

RomiChess is famous for its learning approach ...

Learning
As explained by Michael Sherwin, RomiChess uses two types of learning [5] :

1. Monkey see Monkey do. Romi remembers and incorporates winning lines regardless of which side played the moves into the opening book and can play them back instantly up to 180 ply if the stats for that line remain good.
2. Pavlov's dog experiments adapted to computer chess. Each sides moves are given a slight bonus if that side has won and the other sides moves are given a slight penalty. So, good moves can get a slight penalty and bad moves can get a slight bonus, however, through time those are corrected. These bonus/penalties are loaded into the hash table before each move by the computer. If Romi is loosing game after game then this will cause Romi to 'fish' for better moves to play until Romi starts to win.

Also against Glaurung 2 in a series of ten matches using Noomen's ten position opening suite Romi went from 5% in the first match to 95% in the tenth match. AFAIK RomiChess was the first engine to successfully demonstrate the power of deep RL. So while this is about your book I do believe RomiChess although not NN deserves a mention in the introduction.

Thanks
Mike

The question is if Romi can beat stockfish in a series using Noomen's ten position with learning or maybe after enough games it is going to get only draws.

I am not sure what are the Noomen's ten positions but I will not be surprised if Stockfish get 100% draws against itself in rapid time control from these positions(I know that Stockfish almost never win against itself in Mark Young's matches and for example Stockfish14 won against Stockfish13 3-0 with 997 draws).

https://prodeo.actieforum.com/t445-stoc ... 1000-games

SF using more than 1 thread is very non deterministic making it necessary to play millions of games for Romi to show any progress. I do not remember if I have tested Romi's learning against single threaded SF. I guess that I could do that and report back.