lkaufman wrote: ↑Tue May 04, 2021 5:53 pmWe believe that the MultiPV search is now more effective than the Stockfish MultiPV search, and that if MultiPV is set to more than 4 Dragon 2 is now the favorite against Stockfish 13 on typical modern hardware.
Congrats with the new Dragon in the first place. I want to propose an idea that could reasonable well proof which engine is best with MultiPV=2, the next best move so to say.
8/2pr1kp1/1nR2p2/p2r3p/P2R4/1PP2NP1/6P1/2K5 b - - c0 "g5=10, Ke8=4, Ke7=0, Rxd4=0";
2rr2k1/pp1nppbp/q1np2p1/8/2P1PP2/2N1BQ1P/PPR3P1/1B1R2K1 w - - c0 "Bf2=10, b3=5, Rcc1=8, Kh1=7";
r3k2r/pp1qnppp/1b1p1n2/2p1p3/P3P3/2PP1N1P/1P1NQPP1/R1B1K2R w KQkq - c0 "Nf1=10, Nc4=8, O-O=6, Nh4=6";
The speedy rating list rated engines pretty well using 40,000-80,000 of this type of positions. If engines would have an option to play the next best move you can rate them.
90% of coding is debugging, the other 10% is writing bugs.
Raphexon wrote: ↑Fri May 07, 2021 10:28 pm
Easy to increase multipv strength.
Just output bogus values for the other moves and only search best move.
No need to lie, just output whatever is in the hash for them without additional search if you like.
On the other hand, MCTS engines seem to be able to do multi-pv more cheaply than standard alpha-beta.
In addition, some of the best problem solvers are using multi-pv=2 under the covers to increase tactical strength.
Probably won't help strategic or quiet moves (which is, after all, most moves) but I don't think we can be certain until the experiment is run.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
Raphexon wrote: ↑Fri May 07, 2021 10:28 pm
Easy to increase multipv strength.
Just output bogus values for the other moves and only search best move.
It does not work for this thread
8/2pr1kp1/1nR2p2/p2r3p/P2R4/1PP2NP1/6P1/2K5 b - - c0 "g5=10, Ke8=4, Ke7=0, Rxd4=0";
If your multi-pv contains g5 and bogus moves you get only 10 points
If your multi-pv contains g5 and Ke8 you get 14 points.
I do not claim that Ke8 is the second best move and it seems that stockfish prefer a different move as second best move but the idea is clear assuming that you have a good moves in the epd file.
