An alternative to uho?

[peter]

The competition has picked up a bit and Minic was able to hold Dragon to two draws. I did just notice that Zahak was able to rope-a-dope Stockfish to two draws. Very nice.

At the moment, leela is climbing through the foothills. I expect it will do even better than Dragon and the trailing Stockfish.

Very interesting.

After 36 games of LC0, SF and Dragon holding on 42, Dragon's still leading, SF and LC0 having same performance so far:



https://tcec-chess.com/

[Graham Banks]

Perhaps balanced, but sharp or complex lines are best.
I'd prefer that rather than artificially induced chess with crap opening lines.

[Rebel]

Komodo is the better killer against lower rated engines, remember the odd matches and how bad SF did.

[Chessqueen]

The competition has picked up a bit and Minic was able to hold Dragon to two draws. I did just notice that Zahak was able to rope-a-dope Stockfish to two draws. Very nice.

At the moment, leela is climbing through the foothills. I expect it will do even better than Dragon and the trailing Stockfish.

Very interesting.

After 36 games of LC0, SF and Dragon holding on 42, Dragon's still leading, SF and LC0 having same performance so far:



https://tcec-chess.com/

Why you do NOT talk to the people or team behind TCEC instead, they probably would answer that this type of openings get them the most TCEC followers regardless, if they are using the best openings to test top engines or NOT. I always was and I am still in favor of programmers to provide their own openings, since it would be considered part of the entire engine package

NOTE: Or even better try to convince the TCEC Team to use the chess324 positions which Larry Kaufman just invented?

[Uri Blass]

Komodo is the better killer against lower rated engines, remember the odd matches and how bad SF did.

Being better at odd matches has nothing to do with being better at beating weaker opponents.

If an engine play random moves when it is a rook down(because it evaluate every move as losing with the same evaluation function)
then it is going to be very weak in rook odd matches but it has nothing to do with beating weaker engines when you are never rook down against them in the first place.

If you want to make your engine weak at odd matches then simply change your evaluation to return the same score for every position when it returns more than 5 pawns against yourself.

Stockfish did something similiar at least for big difference in the evaluation function.

The following position is evaluated by stockfish as more than 21 pawns advantage for white.
[fen]1nb1kbn1/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQ - 0 1[/fen]

The following position that is obviously relatively better for white because white want to trade pieces is evaluated as less than 21 pawns for white
[fen]4k3/pppppppp/8/8/8/8/PPPPPPPP/R2QK2R w KQ - 0 1[/fen]

They destroyed the evaluation function for simplification even before implementing NN's because the developers are not chess players who care about quality of analysis but about simplification of the code and if there is a simplification that destroy the analysis in winning position but does not cost elo points they do not care so they removed something that they did not need to remove from SF classical evaluation.

[Lazy_Frank]

They destroyed the evaluation function for simplification even before implementing NN's because the developers are not chess players who care about quality of analysis but about simplification of the code and if there is a simplification that destroy the analysis in winning position but does not cost elo points they do not care so they removed something that they did not need to remove from SF classical evaluation.

Well said.
Let's hope SF developers doesn't see fishtest as just a programming competition.
After all fishtest was created to build the engine that play's the chess and with time developers take more attention to game play and moves quality too.

[Uri Blass]

I think one patch that destroyed stockfish's evaluation is the following patch.
Note that I am not against lazy evaluation but you need to do it correctly(meaning doing it only when the evaluation in the engine output is not sufficiently large and one side is not clearly winning that is usually the case).

If the engine shows 0.00 or a small score in the output then using a lazy evaluation is a good idea because a score of +17 pawns or +19 pawns in the evaluation is not important to calculate but if the engine shows 17.00 pawns in the output then using a lazy evaluation is a bad idea.


Author: mstembera
Date: Fri Jul 17 20:24:02 2020 +0200
Timestamp: 1595010242

Multiple lazy stages.

An extension of the lazy eval idea: when the score is sufficiently large
we now skip more granular parts of the eval.

Inspired by an original patch by Moez Jellouli
https://tests.stockfishchess.org/tests/ ... 3532894529
Credit to him!

STC https://tests.stockfishchess.org/tests/ ... 3532894924
LLR: 2.95 (-2.94,2.94) {-0.50,1.50}
Total: 13504 W: 2684 L: 2472 D: 8348 Elo +5.45
Ptnml(0-2): 229, 1496, 3111, 1666, 250

LTC https://tests.stockfishchess.org/tests/ ... 353289495b
LLR: 2.94 (-2.94,2.94) {0.25,1.75}
Total: 31312 W: 3926 L: 3677 D: 23709 Elo +2.76
Ptnml(0-2): 185, 2773, 9509, 2986, 203

closes https://github.com/official-stockfish/S ... /pull/2814

[dkappe]

After 42 games, lc0 trails SF. I would have expected a better performance, especially given the extra gpu juice at TCEC.

Now I’d expect SF’s search advantage to gradually make itself felt. (Incidentally, the handful of elo that contempt yields for Dragon can’t explain this performance. But it’s all SSS for now.)