How to handle simultaneously different evaluations?

[Reinhard Scharnagl]

Hi all,

you will know that mating and remis evaluations are different from positionally enriched material balance values. The remis value is not distinguishable from a zero material balance. How do you work with this circumstance? E.g. by forcing non-zero material evaluations?

Reinhard.

[Dan Honeycutt]

Hi Reinhard:

What is "remis". I don't know the term.

Dan H.

[Volker Pittlik]

Hi Reinhard:

What is "remis". I don't know the term.

Dan H.

Draw.

Volker

[Dan Honeycutt]

Hi Reinhard:

Now that I know what remis is (thanks Volker) if you are looking to determine draw in n there was a recent CCC thread:

http://www.talkchess.com/forums/1/message.html?397117

I've been tinkering with the idea of 2 evaluation scores, a traditional centi-pawn score for material/positional advantage and a probability score based on piece mix, half move clock or other heuristics. I only plan to do this for endgame situations. The general idea is if the centi-pawn score is good for one side but their winning probability is low I should pull the score closer to a draw. Conversly, if the centi-pawn score is about equal but one side has better winning probability I adjust the score to reflect that. No idea if this will do any good.

Dan H.

[Reinhard Scharnagl]

Hi Dan,

I intend to do the task with ONE evaluating scheme. Because I think it will make no sense to have two different transposition table evaluation entries for identic positions.

I hope to be right that when expanding a node for evaluation I have to switch to the evaluation using only final values like being mate, set mate and draw (first both with different levels) then when a mate will apear.

Reinhard.

[Richard Pijl]

The Baron has 3 values reserved for draws: 0.01, 0 and -0.01 representing a draw with more material, draw with equal material and draw with less material. In the evaluation function the positive scores are increased by 0.01, negative scores decreased by 0.01. 0 scores are 0.02 from the side to move perspective.
Richard.

[Reinhard Scharnagl]

Hi Richard,

nice to meet you here again!

If I understand you correctly, you are also avoiding to use 0.00 to encode balanced material, but in a modified way.

Reinhard.

[Richard Pijl]

If I understand you correctly, you are also avoiding to use 0.00 to encode balanced material, but in a modified way.

Exactly. Basically I'm mapping 0 and 0.01 both to 0.02 (from side2move perspective), but I don't think that is a big problem.

The reason that I have 3 draw values is to prevent a random choice between drawing moves. You can have a drawing move that wins a piece or one that loses a piece. Especially when playing humans (or 'puters without EGTB's), you want to play the move that wins the piece (or in any case doesn't lose one) as maybe the opponent not able to find the correct continuation which can result in a win later.
Richard.