Moderators: hgm, Rebel, chrisw
From Dragon readme file:
Thanks for the info, appreciate itcc2150dx wrote: ↑Mon Nov 23, 2020 2:42 pmFrom Dragon readme file:
"A positive Contempt is used by the computer’s evaluation to discourage (or encourage if negative) piece trades, blocked pawn structures, and repetition draws early in the game."
"The default value 8 should help against weaker opponents. Against near-equal opponents e.g. (Stockfish 12/dev or Lc0) set Contempt to 0 for best results."
P.S. The second quote has been slightly modified.
Settings that affect eval, such as Dynamism or King Safety, will have little effect in Dragon mode; they will only matter when regular eval is used, which in default mode is only in positions where it is unlikely to matter. The search parameters all work normally though.Nordlandia wrote: ↑Mon Nov 23, 2020 4:18 pm I tried this unique position for Dragon as black against SF and Lc0. It looks like white has good compensation for a couple of pawns. The compromise is that blacks are not allowed to castle.
So far, the games ends as draw. I was wondering if any other settings apart from default makes it doable for Dragon to win as black. For example if i increase/lower dynamism to make Dragon play more conservative or riskier with material advantage. What is your suggestion here. Is it likely that there are some exceptional settings that can be discovered ? comparable to "Kiudee-setting"
[d]rnbqkbnr/pppppppp/8/8/8/8/1PPPPPP1/RNBQKBNR w KQ - 0 1
Furthermore, Larry Kaufman based his claims on a sample size of one (a blitz game) which is simply not enough evidence to conclude anythung. In fact, extensive testing by Stefan Pohl show Dragon MCTS to be 180 elo worse than Dragon AB, around the level of Stockfish 9.Alayan wrote: ↑Tue Nov 10, 2020 8:42 pmThis marketing claim "forgets" to acknowledge that this increased strength of the first move is achieved by reducing how deep the engines looks at lines beyond the first one. Indeed, the ability to report multiple PVs with eval, and dedicating similar resources to explore each of the lines, are separate concepts.lkaufman wrote: ↑Tue Nov 10, 2020 7:42 pm When both engines were set to use MultiPV=4 for standard chess, Dragon MCTS was able to defeat SF12 decisively in a 3600 games blitz match, so we believe that on most computers it is the world's strongest engine when looking at four or more lines of play at once.
Play with MultiPV=4 only reflects the quality of the first move, not that of the 2nd, 3rd or 4th, and is a fundamentally flawed way to evaluate how useful a MultiPV implementation is.
The increased quality of the first move in Komodo (and Leela) MCTS comes at the cost of a lowered quality and reliability for the moves, PV and evals of the 2nd and further lines come from smaller search trees. Turning MultiPV on in these engines doesn't increase the likelihood that they discover an initially disliked move is much better than first thought.
Most people that do use MultiPV for analysis expressed strong opposition to this trade-off when reducing 2nd and further PV lines was considered for Stockfish, though in some circumstances (moves that are much worse than the main move) it can be useful.
For deep analysis, multiPV isn't going to be used on positions with one clear forced move anyway, and the effective time-handicap that makes KMCTS look good at blitz matter much less while the search/eval deficit remain, making Stockfish the best MultiPV engine for e.g. correspondence chess.
??? My sample size was 3600 games, not one game! I don't follow your logic here at all. We certainly don't claim that Dragon AB is superior to Dragon MCTS for Multipv (3 or more) analysis, but it is certainly superior for analysis with no MultiPV. MultiPV cripples normal engines like Dragon AB and Stockfish, that's why Dragon MCTS can win a long match from SF 12 if both use MultiPV of 4. If they don't use MultiPV, then Dragon MCTS is still stronger than Komodo 14.1 in standard mode (as Stefan Pohl confirmed), but it is weaker than SF11, never mind 12. Our testing is not contradicted by Stefan Pohl's results, we're talking about two very different uses of the engines.Madeleine Birchfield wrote: ↑Mon Nov 23, 2020 9:18 pmFurthermore, Larry Kaufman based his claims on a sample size of one (a blitz game) which is simply not enough evidence to conclude anythung. In fact, extensive testing by Stefan Pohl show Dragon MCTS to be 180 elo worse than Dragon AB, around the level of Stockfish 9.Alayan wrote: ↑Tue Nov 10, 2020 8:42 pmThis marketing claim "forgets" to acknowledge that this increased strength of the first move is achieved by reducing how deep the engines looks at lines beyond the first one. Indeed, the ability to report multiple PVs with eval, and dedicating similar resources to explore each of the lines, are separate concepts.lkaufman wrote: ↑Tue Nov 10, 2020 7:42 pm When both engines were set to use MultiPV=4 for standard chess, Dragon MCTS was able to defeat SF12 decisively in a 3600 games blitz match, so we believe that on most computers it is the world's strongest engine when looking at four or more lines of play at once.
Play with MultiPV=4 only reflects the quality of the first move, not that of the 2nd, 3rd or 4th, and is a fundamentally flawed way to evaluate how useful a MultiPV implementation is.
The increased quality of the first move in Komodo (and Leela) MCTS comes at the cost of a lowered quality and reliability for the moves, PV and evals of the 2nd and further lines come from smaller search trees. Turning MultiPV on in these engines doesn't increase the likelihood that they discover an initially disliked move is much better than first thought.
Most people that do use MultiPV for analysis expressed strong opposition to this trade-off when reducing 2nd and further PV lines was considered for Stockfish, though in some circumstances (moves that are much worse than the main move) it can be useful.
For deep analysis, multiPV isn't going to be used on positions with one clear forced move anyway, and the effective time-handicap that makes KMCTS look good at blitz matter much less while the search/eval deficit remain, making Stockfish the best MultiPV engine for e.g. correspondence chess.
http://talkchess.com/forum3/viewtopic.php?f=6&t=75900
https://www.sp-cc.de
Following Larry Kaufman's logic with this evidence would mean that Dragon AB is far better than Dragon MCTS for analysis.
Why doesn't MwYoung start a 200 games match like you did with Stockfish normal but against MCTS vs Stockfish-NNUE in FRC?mwyoung wrote: ↑Fri Nov 13, 2020 6:36 pmAt 5m+5s on a 32 thread 2950x. Dragon is doing much better then 80 Elo. Very good for the first release of Dragon.
Mark L. asked my to change the default setting for hash memory from default. I do not know if this changed the results. Default was 64, and was changed to 512. The suggested best setting for this many threads.
Here is the results playing the best Stockfish.
Dragon by Komodo Chess vs Stockfish 101120 (TC=5m+5s)
With 198 games played so far. Dragon vs Stockfish 101120 +6 =172 -20. 46.5% TP = -25 Elo.
Live Stream
Better yet...how can only 6 wins vs 20 losses = a mere -25 elo difference?Chessqueen wrote: ↑Tue Nov 24, 2020 5:53 pmWhy doesn't MwYoung start a 200 games match like you did with Stockfish normal but against MCTS vs Stockfish-NNUE in FRC?mwyoung wrote: ↑Fri Nov 13, 2020 6:36 pmAt 5m+5s on a 32 thread 2950x. Dragon is doing much better then 80 Elo. Very good for the first release of Dragon.
Mark L. asked my to change the default setting for hash memory from default. I do not know if this changed the results. Default was 64, and was changed to 512. The suggested best setting for this many threads.
Here is the results playing the best Stockfish.
Dragon by Komodo Chess vs Stockfish 101120 (TC=5m+5s)
With 198 games played so far. Dragon vs Stockfish 101120 +6 =172 -20. 46.5% TP = -25 Elo.
Live Stream