SPCC: Testrun of BrainFish 171104 finished

[pohl4711]

Testrun of BrainFish 171104 finished.

Long thinking-time tournament updated.


http://www.sp-cc.de

(Perhaps you have to clear your browsercache or reload the website)

[ernest]

Just curious, Stefan,
when you test Brainfish (or asmFish) + Cerebellum, what book do the opponents get, and how do you reverse the openings ?

[pohl4711]

Just curious, Stefan,
when you test Brainfish (or asmFish) + Cerebellum, what book do the opponents get, and how do you reverse the openings ?

The other engines are using no books. Each game starts with a position of the HERT opening-set.
BrainFish and asmBrainFish are using Cerebellum, but Cerebellum is built only upon engine evaluations (Stockfish). It can be used for analysing positions and trees of trees of variations. Additionally, it can be used as an opening library for chess engines.
Does the Cerebellum Library simply consists of a lot of calculated Postions? No, the Cerebellum Library does a recalculation of every score in the Library. That recalculation is done with a graph Algorithm, so that every Score in a node is consistant with the score of it's best successor (backwards minimaxed node-evaluation). Furthermore the Cerebellum Library is able handle position repetitions by with re-visiting some nodes and then playing different moves.

So, the Cerebellum-Library has much more similarities to a huge file of persistent-hash data, all created by Stockfish itself, than to an opening book.
So, when Stockfish uses Cerebellum, it uses only its own calculation-data. That means, that everything, what BrainFish and asmBrainFish are playing, comes from the Stockfish-Engine.
And with my testruns, Thomas Zipproth and me, want to measure, how much Elo can Stockfish gain, when using its own pre-calculated data/moves, compared to Stockfish without using such data. And the played games, using common openings (HERT), help Thomas, to improve Cerebellum.
These testruns of BrainFish and asmBrainFish are experimental. And I dont have a ratinglist. So, it doesnt matter, if using Cerebellum is “fair“ or not. And other engines could use own pre-calculated data. They doesnt do it - until now, but who knows, what is coming in the future?!

Stefan