Winboard Forum

Hi everyone,

I just want to ask if there is an experiment out there where they disabled

the eval_draw component of Fruit's evaluation function and test it whether

it is still as strong as before the change. I'm just wondering, it might be

one of the factors that makes Fruit stronger than most of the chess

engines. In my opinion, it enables Fruit to avoid entering a drawish

position if it is ahead and to strive for that drawish position when it is

losing. This would result to Fruit minimizing the number of its losses and

therefore increasing its strenght. Just a wild idea but it might be true.

I am not aware of such a test, but it is true that Fruit wins many of its games in the endgame or in the transition from middlegame to endgame.

Robert

I am not aware of such a test, but I share in some parts of your theory.

Fabien has a very good idea to discover as many 'absolute' scores as possible in the tree, for these will cause cut-offs as well as be 'absolutely' accurate and therefore good for the evaluation

Hello Edsel,
I do not think it is a wild idea at all, although I'm not aware of any serious tests that have proven it's effectiveness. It does make plenty of sense to me. (of course my engine is not very good either).

I do not think that the eval_draw is any sort of key to Fruit's strength.
Every program with bitbase access gets not only the same information, but better information (because some of Fabien's draw estimates are only good guesses).

If it were a key to such a stupendous advantage, then engines like Patzer, Yace, KnightD3 and Scorpio would be equally strong because they have the same information (only a teeny bit better).

Also, tablebase enabled programs would have the same power if you put the 6 or so tablebase files needed to emulate Fabien's draw guesses on Ramdisk. So that would be a way to find out. I would be very surprised if you were to see an Elo gain of more than 5 points.

Well maybe you're right, but it might be that those programs you

mentioned have inferior search function, and their evaluation weights are

untuned not just like Fruit. My point is, in Fruit's point of view, I guess it

would decrease it's elo rating if it disabled it's eval_draw component

especially if it's opponent are those commercial programs.