hgm wrote:I think that is a very bad thing. If Stockfish reports wrong scores, it should be fixed in Stockfish.
Why do you think Stockfish is wrong? I think it's just that the users are adapted to Houdini's evaluation now. Fixing users would seem to be the job of the GUI.
Well, if they differ, one of the two must be wrong. It should not be difficult to figure out which of the two. There are hundreds of engines that do it right. I use neither Houdini nor Stockfish, so I wouldn't know.
There is a clear standard for how scores should be reported, however (centiPawn), both in WB protocol and UCI. If engines report scores that should be multiplied by a fixed factor to be meaningfully compared with other engine scores, these engines are non-compliant, and should be fixed. Encouraging such non-compliance by letting the GUI fix it on an engine by engine basis is helping to create chaos.
If there are users that rather see the score in full Pawn units, or milliQueens, or 1/256 Pawns, then doing that conversion for all engines alike would be a GUI task. But users should not be bothered calibrating individual engines.
I don't think there is a standard involved ? This is simply the engine's evaluation of the position. As it happens, Stockfish evaluations are a little different from say Houdini. And for example, I think Gaviota is usually a little more optimistic than other engines.
hgm wrote:Well, if they differ, one of the two must be wrong. It should not be difficult to figure out which of the two. There are hundreds of engines that do it right. I use neither Houdini nor Stockfish, so I wouldn't know.
There is a clear standard for how scores should be reported, however (centiPawn), both in WB protocol and UCI. If engines report scores that should be multiplied by a fixed factor to be meaningfully compared with other engine scores, these engines are non-compliant, and should be fixed. Encouraging such non-compliance by letting the GUI fix it on an engine by engine basis is helping to create chaos.
If there are users that rather see the score in full Pawn units, or milliQueens, or 1/256 Pawns, then doing that conversion for all engines alike would be a GUI task. But users should not be bothered calibrating individual engines.
I do not think that it is clear what is the meaning of a pawn with all the positional bonuses that programs have in their evaluation for mobility passed pawns and other things.
difference in a score between programs can be not because of different material evaluation but because of different positional evaluation.
The side that has material advantage often has more passed pawns or better mobility so practically the evaluation may be bigger than the material difference.
hgm wrote:Well, if they differ, one of the two must be wrong. It should not be difficult to figure out which of the two. There are hundreds of engines that do it right. I use neither Houdini nor Stockfish, so I wouldn't know.
There is a clear standard for how scores should be reported, however (centiPawn), both in WB protocol and UCI. If engines report scores that should be multiplied by a fixed factor to be meaningfully compared with other engine scores, these engines are non-compliant, and should be fixed. Encouraging such non-compliance by letting the GUI fix it on an engine by engine basis is helping to create chaos.
If there are users that rather see the score in full Pawn units, or milliQueens, or 1/256 Pawns, then doing that conversion for all engines alike would be a GUI task. But users should not be bothered calibrating individual engines.
SF reports scores in centipawns directly from the evaluation. Houdini actually scales it's internal scores into win probabilities. Both are completely valid IMO.
Graham Banks wrote:
Matthias - have you managed to implement sudden death games for tied knockout matches?
If you haven't, I'd really appreciate it if you could.
The other feature that would be useful would be the evaluation given in brackets after each move in the displayed pgn.
I've sent you several emails during the last few months but you haven't replied. My top requests:
1: Will you look to improve the pairing algorithm for round robins?
2: Is it possible to get the full PV instead of the 12 move limited version?
Best,
Martin
I don't think you earnestly expected me to reply to any of them. Why should I ?
I had a highlighting colour in your chat in Stage 1 of your Season 2 TCEC.
From Stage 2 onwards, you disqualified me from having a green colour in your chat. Even when I specifically asked for a green colour, you refused giving it, pointing me to a rule you had made that permits only engine authors to have a highlighting colour.
I have so far invested dozens of hours tweaking ChessGUI to meet your TCEC demands. Why should I continue doing the dirty work back-stage when a mere green colour in the chat is too much to honour me with ?
I understand that no other GUI is able to replace ChessGUI in TCEC at the
moment, so use it till alternatives become available.
or
or
?
