Leela Zero

Vous rêvassez encore des performances d’AlphaZero ?  Arrêtez tout de suite!  Une âme charitable a décidé d’utiliser les mêmes méthodes d’apprentissage qu’AlphaZero et de les appliquer à StockFish!  Et voilà, Leela Zero est née!  Un fantastique projet open source qui sollicite votre participation! Vous pouvez même affronter la plus récente version de Leela Zero en ligne ici.

KNNN vs K

Finalement, les nombreuses (et parfois complexes) explications suite à mon post (concernant la finale roi et 3 cavaliers contre roi) sont ici.

The 3 knights

I was just curious, nothing more.  I was asking myself how hard is it to checkmate with 3 knights so I fired up Stockfish (without any tablebase) and was shockingly surprised to see the search results bump up and down showing different « distance-to-mate » !  How can a search at 41 plies deep jump from 43-35-36-28-25-33 plies to mate?!?!?  Does anyone have an answer?

I have checked with Lomonosov tablebases (mate in 17), Syzygy (DTZ 32) and Nalimov tablebases (mate in 17 as well) so it looks like it’s a mate in 33 plies.

So I’m wondering why Stockfish, at 41 plies deep, is not able to stick to a mate-in-33 principal variation all along starting from ply 33 ?

Thoughts?  Explanations?

Endgame tablebases online

Kasparov vs The World, 1999, after 58.g6

You don’t have endgame tablebases installed on your computer?  Don’t worry, they’re available online here and here!

Humour (120)

Chess arguments…

Finally!

For those interested, I have released the very first version of my database of chess games, the CGR database!  All details here!

The Chess Games Repository

What do you need to build a solid opening book for a chess engine? And what do you need to study chess openings seriously?

Games.  Lots of games. Lots of games by strong players. So my long-dead project revives again!

The Chess Games Repository!