TalkChess.com

This is very good.
"Make Your Own Neural Network" by Tariq Rashid.
A good starting point. It doesn't cover Convolutional NN but all the basics are there.

Hello, I'm having this idea about a chess engine that learns from genetic algorithms. Basically, every generation plays itself and we select the winner for the next generation. Has this been done before? Would it work? Any ideas? iCE is a good and successfull example. If I recall correctly, it uses...

Hello, I'm having this idea about a chess engine that learns from genetic algorithms. Basically, every generation plays itself and we select the winner for the next generation. Has this been done before? Would it work? Any ideas? iCE is a good and successfull example. If I recall correctly, it uses...

(this->type == typeExact) || (this->type == typeAlpha && this->eval <= alpha) || (this->type == typeBeta && this->eval >= beta) Thanks for your inputs. So you suggest to use the TT move and return a score with it as soon as it is an exact score, even at a PV node ? Don't forget the depth condition ...

!fromPV || (ttt.t_type == Transposition::tt_exact && ttt.score > alpha && ttt.score < beta) || (ttt.t_type == Transposition::tt_alpha && ttt.score <= alpha) || (ttt.t_type == Transposition::tt_beta && ttt.score >= beta) ) ) Moreover, I used to not use the TT move directly (i.e. return a score immed...

Also, in a 1 ply search, I'm selecting the move with the best LOS, not the move with the best percentage. Have you tried selecting the move with the lowest lose percentage? Alpha Zero gave the impression of going for positions that first and foremost, it could see no way of losing, so it may explai...

Isn't it a kind of monte-carlo playout? Since the Alpha-zero success I've been fooling around the idea of the playouts to evaluate the board. What I found was that in the random game you can't select the moves with absolute randomness. you have to guide the moves with some evaluation (I'm using quie...

Manik is convinced that there is a bug in the move ordering code of Dorpsgek due to having a comparatively low fail-high on first move rate (I ran a self-play game to depth 8 and got an average of 84.8%), and after a couple of months of not being motivated, I thought I'd ask the CCC. My move orderi...

Sorry, another thread posted it yesterday:

http://www.talkchess.com/forum/viewtopic.php?t=65481

Looks like a milestone in IA. https://deepmind.com/blog/alphago-zero-learning-scratch/ They say: Previous versions of AlphaGo initially trained on thousands of human amateur and professional games to learn how to play Go. AlphaGo Zero skips this step and learns to play simply by playing games agains...