Computer evalutation variance based on interface.

View previous topic View next topic Go down

Computer evalutation variance based on interface.

Post by Bilbo on July 8th 2009, 1:08 pm

This is just a weird thing I've noticed with my own setup and wondering if anyone experiences this or has an explanation.

When computer analysing my games I initially do it in Fritz but using the latest Rybka engine.

When playing games back however I play them in chessbase 10 as I prefer the layout, font and colour is more attractive, but I have the Rybka engine running whilst I play through them in the background.

However the engine gives slightly different evaluations in Chessbase 10 than in Fritz, even though I'm using the same Rybka engine in both.

For example my game from Tuesday was a Queens Gambit Declined in which my opponent play the usual Nd5 to force some exchange with Bxe7 Qxe7 etc. As my opponent had earlier wasted a move with Nb6 however I felt it was better not to exchange here but to play Bg3 (he had earlier played h6) and save my bishop as his queenside development was virtually impossible with me overprotecting e5 to prevent a freeing pawn advance, taking away c7 for his queen and with his Knight on b6 his light squared bishop was stuck.

After a few seconds Rybka agreed with me, changing its initial assessment from slightly favouring Bxe7 to about equal in the first couple seconds to then favouring Bg3 with the appox score 0.59 in my favour after Bg3 and 0.49 after Bxe7, which is what I myself would have expected.

When playing through the same game today though in Chessbase 10 but still using Rybka as the engine it prefers Bxe7. I left it for a couple minutes and it still has a preference for Bxe7 over Bg3 by about 0.05 of pawn.

Not a big deal but I'd like to know why the discrepancy when I'm using the exact same chess engine but in two different programs (Fritz 11 and Chessbase 10)?

Bilbo
International Master
International Master

Posts : 269
Join date : 2009-04-18

View user profile

Back to top Go down

Re: Computer evalutation variance based on interface.

Post by HangingKing on July 8th 2009, 4:46 pm

It's possible, chess engines today have a lot of configurable parameters, and the default values chosen by the GUI (the interface program) are likely to be different.

Check for "outlook" and "contempt" values assigned to Rybka engine in both environnement, if they are different don't search further.

Personnaly i use Arena to run evaluations.

Another thing with Rybka is how you analyse the game, you will get different results if you analyse a FEN position and if you analyse a PGN from the start, also if you try variants and come back to the initial position.

I still use version 2.3.2a for analysis, while version 3.1 gives smaller variations and is more subject to lunatic evaluation.
I often use Nejmet engine too for evaluating, because it shows more variations than Rybka.

__________________________________________
Never give up, never surrender.
avatar
HangingKing
International Master
International Master

Posts : 371
Join date : 2009-04-21

View user profile

Back to top Go down

Re: Computer evalutation variance based on interface.

Post by Bilbo on July 8th 2009, 4:55 pm

haha I didn't understand a word of that Very Happy

But as long as its a common thing I'm not too worried, just curious as to why games were evaluated slightly different between the two interfaces.

The engine I use is called Rybka 3 Human1-cpu

Bilbo
International Master
International Master

Posts : 269
Join date : 2009-04-18

View user profile

Back to top Go down

Re: Computer evalutation variance based on interface.

Post by fanat on July 8th 2009, 6:42 pm

What Hanging King is trying to say is that Rybka might have different engine parameters or setting set in Fritz interface and Chessbase. So, it might cause the evaluations to be different.
avatar
fanat
National Master
National Master

Posts : 172
Join date : 2009-04-14

View user profile

Back to top Go down

Re: Computer evalutation variance based on interface.

Post by HangingKing on July 30th 2009, 12:25 am

I have found an interesting chess engine.
It is free and work in Arena (i did not tested other interface).

http://transversale.fr/Arion/Arion.htm

In fact i was tired of loosing, em... being crushed without mercy, while Rybka, Crafty or Fritz were thinking only 1 or 2 plys forward and answered instantaneously to all my moves.

Well, despite some options to weaken this engines, this is too basic, and the evaluation function is so strong even with no deep calculation.
These are perfect engines for post-mortem analysis but not for training play, even weakened to the ground.

And I found Arion, which as the cool fun option !
(as described on the html page above).

The engine adapt dynamically in the course of the game to your own strength. I tried it and found the games very appealing.

The fact that the moves are not ridiculously weak, nor so strong that they let you no chance, this kept me in the maximal concentration trying to raise my level of play.

I suggest you give a try to this engine. :-)

__________________________________________
Never give up, never surrender.
avatar
HangingKing
International Master
International Master

Posts : 371
Join date : 2009-04-21

View user profile

Back to top Go down

Re: Computer evalutation variance based on interface.

Post by Sponsored content


Sponsored content


Back to top Go down

View previous topic View next topic Back to top


 
Permissions in this forum:
You cannot reply to topics in this forum