Not logged inRybka Chess Community Forum
- - By Rebel (****) Date 2019-09-23 17:42 Upvotes 1
http://rebel13.nl/misc/sim2019.html

Find the similarities between 2011 and 2019

:eek:
Parent - - By Labyrinth (*****) Date 2019-09-23 22:51
Fascinating. I can't help but wonder if increasing strength is headed towards some sort of 'norm' though, like a traditional AB engine past a certain Elo will have greater and greater similarity as the same techniques are required.

One thing appears to be for sure though: if you have an open source software that does something very well, parts of it are bound to show up in closed source programs regardless of licensing!
Parent - By Rebel (****) Date 2019-09-25 18:37
Note that this is about measuring the similarity of the evaluation function in chess programs. You (likely) will remember the ICGA Rybka-Fruit case when it was for 90% about the similarity of the evaluation with Fruit.
Parent - - By Uly (Gold) Date 2019-10-08 10:10

> I can't help but wonder if increasing strength is headed towards some sort of 'norm' though, like a traditional AB engine past a certain Elo will have greater and greater similarity as the same techniques are required.


You can get an Stockfish to play in a CPU at about the same strength of a Leela playing in a GPU, and both will play a very different chess.

I think that if something happened radically differently about chess engine programming early on, a completely different course would have taken place, and probably engines would be as strong as today (or even stronger?) while playing a completely different chess than they do now.

To see what I mean, ask a newbie programmer to make a chess engine, but she has to come up with her own code from scratch. Assuming she doesn't know about Alphabeta or Minimax, she'll probably make an engine that will play very unlike anything that exists today, and if she could continue working on it, she could improve it until you get something really strong and really different, there's just no incentive for reinventing the wheel (people wanting to program their chess engine are researching what people are doing and what seems to be working before even starting), but the reason some similarity exists is that these programs are applying the same ideas at the base. NNs are not, and that's why they're not similar.
Parent - By Labyrinth (*****) Date 2019-10-09 00:06
I know about the different chess, but I mean programming implementations for a strong, traditional style engine like stockfish. I can't help but wonder if there is a norm there, it's going to have to be sort of fast, and there are only so many ways to calculate efficiently. Could build into some algorithmic type positional understanding while keeping the node count low, but I don't think anywhere has gotten anywhere near the top with such an approach. That's why I wonder if there is a norm there. The norm for chess itself is a completely different animal, and may not exist.

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill