Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Support / Is somewhere a documentation file for Rybka 2.x ?
- - By Zsolt Kántor (**) Date 2019-06-12 10:55
For the Rybka free engines like 2.2n2 and/or 2.3.2a is somewhere a in depth documentation file? I know only about the online FAQ for Rybka 2's Parameters page (http://www.rybkachess.com/index.php?auswahl=FAQ+for+v+2.x). But is there a more detailed doc file?

Thanks.
Parent - - By Uly (Gold) Date 2019-06-12 21:47
Nope, that's as detailed as you'll get.

However, if you have an specific inquiry about any of its options, I probably know the answer.
Parent - - By Zsolt Kántor (**) Date 2019-06-13 20:35
Thanks for your support. At the moment I have no specific questions. I get tired to play with the modern UCI engines, so I switched to Rybka 2.2n2.
Parent - - By Uly (Gold) Date 2019-06-13 21:09
Have you tried playing against the Rybka Winfinder version? I think for playing against an engine that's the one that would be the most fun.
Parent - - By Zsolt Kántor (**) Date 2019-06-13 21:31
I read something about Rybka winfinder, but I don't remember exactly what's the main difference between the "normal" versions . . .
What's the purpose of that winfinder thing?

Thanks!
Parent - - By Uly (Gold) Date 2019-06-13 21:46
In general, normal engines play very cautiously, because they assume the opponent will not miss anything that they see. This increases their ELO as against equal opposition that's a right assumption. However, this is not best against weaker opposition, where they'd have missed the best defense and thus, the move a normal engine rejects (because it sees its refutation) would have been best.

What Winfinder does is trying to find a win in a position at all costs, extending such refutations beyond what Rybka would have done, in case further examination reveals a win. It'd go as far as missing an existing win that the opponent has if it plays the moves it plays, because it only cares about winning.

This creates an uniquely shaped tree that produces the most stylistic Rybka of that era (an engine that would play a potentially losing move because it blinds itself against opponent refutations?), though of course weaker than Rybka.

At least, this was what was observed against other engines, I don't recall ever a human playing against it. A human looking for something like this would play against something like Pro Deo with Q3 personality, that has a similar vein though radically different playing style.
Parent - - By Zsolt Kántor (**) Date 2019-06-13 22:37
I just found a post where you shared in a 7zip file all the free for use rybka engines, there was the Winfinder also. I installed it. At a first glance it is not so aggressive as Rybka 2.2n2, it feels to me it has a smoother playing style . . .
Parent - - By Uly (Gold) Date 2019-06-14 10:52
Was this Rybka Winfinder or Rybka Winfinder 2.2?
Parent - By Zsolt Kántor (**) Date 2019-06-14 15:19
In the package is WinFinder 1.0 and 2.2 mp.
Parent - - By Zsolt Kántor (**) Date 2019-06-14 16:29
I found a readme file for WinFinder 1.0 (RybkaWinFinder10readme.rtf). I searched for 2.2 but nothing found. Do you know a newer readme/doc for the WinFinder engine?
Thanks.
Parent - By Uly (Gold) Date 2019-06-14 17:03
No, but Winfinder 2.2 is just Rybka 2.2 with different search so what works for Rybka works for Winfinder.
Parent - - By Labyrinth (*****) Date 2019-06-14 02:29
You should make a list somewhere of the engines you're familiar with and their various styles, like which engine is the most insane/aggressive? Which the most defensive and positional? etc
Parent - - By Uly (Gold) Date 2019-06-14 10:49
What I keep coming back to is the idea to create a method that can be computed automatically to measure those styles, and get subjectivity out of the way. The idea would be to give big bonuses to major pieces for minor pieces exchange (an engine that swaps their queen for a pawn like nothing would be greatly rewarded), pieces hanging, sacrificial king attacks and so on.

What is difficult is that we also want the moves to work, I can easily program a Pro Deo personality that breaks the system and beats everyone in style but loses every game. Matching opponents of same strength doesn't work either, I could create more personalities that lose against that and inflate its "style."

The dream would be to build a formula that computes that and then divides by elo, so we get some "Style/Strength" ratio. And then we'd just change what we measure in style, so we can create some kind of Hexagon with some "Insane/Aggressive/King Hunter/Solid/Defensive/Active" at the tips and then we use this to show an exagon for chess entities. The cool thing about this is that once it's set up, you don't need hundreds of thousands of games to measure style, about 100 games against a wide range of opponents would suffice.

Brendan Norman (ChessCognac) has been dedicating a lot of his time to explore and measure engines' styles, unfortunately, he has been doing so to sell videos about engines' styles. But he might know more about them than me. After all, currently the tip of engines style seem to be occupied by Rodent Personalities, which seem capable of producing chess games where the point is to see something cool and unique happen, and they succeed every time. And I have never even tried Rodent.

But about all this I can say two things:

1. Houdini 6 at Contempt 10 would break any system I device, and I still don't know if some Stockfish derivative with Contempt 200 or something would break it even further. At least they had solved the problem of stronger engines playing with worse style, which is nowhere to be seen.

2. The insanest chess entity is Leela. Kingscrusher at Youtube has been showcasing Leela games, and, man, I continuously see new things that a "style test" wouldn't have included, so how does she even fit? How big should a bonus be if there's only one chess entity that would get that bonus?
Parent - - By Labyrinth (*****) Date 2019-06-14 20:16
I've seen a lot of those Leela videos, it's a pretty creative/original engine in terms of style that's for sure.

What is Houdini with contempt 10 like?
Parent - By Uly (Gold) Date 2019-06-15 03:15 Edited 2019-06-15 03:18

> What is Houdini with contempt 10 like?


Depends :lol:

It's something unique, it seems to detect what side has the advantage, and adjust its style accordingly. So it might start the analysis thinking it has big advantage (+0.60 scores) and suggest some big attack, but if it sees the attack doesn't work it'd quickly switch to something more conservative and a score that shows it's back into some kind of counter-attack mode (score -0.05). In other situations, it would show some cautious line and move choice (+0.19 scores) but once it sees it had nothing to be afraid of, it'd search for an attack, switch to radically different line and unleash the beast (+0.90 scores - though scores don't say much here, other than the other side has no chance to counter-play or it'd not have switched moves.)

It's like many personalities rolled into one, and it adjusts depending on the situation. Most other engines go insane or passive all the way, Houdini 6 knows when to stop, or when to look for something, that's why I praise its style (which... remains style. This isn't the default because most attacks won't work or it'd have been better to not go into the defensive, and of course it can go bipolar switching back and forth between its moves showing score roller coasters, but it'll NEVER show a 0.00 score, opting instead for scores showing what side is most likely to go into attacking mode.)

Back in the times of Stockfish 8 all this was extraordinary. Nowadays, not so much, as Stockfish derivatives can play solid moves and only launch an attack if they find it naturally, Contempt and Dynamic Contempt of Stockfish kind of made using other engines of different styles unneccesary. I don't think I have used Houdini at all in the last 6 months.

Edit - All this with Tactical = 1...
Up Topic Rybka Support & Discussion / Rybka Support / Is somewhere a documentation file for Rybka 2.x ?

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill