Vinvin wrote: ↑Tue May 18, 2021 11:28 pm
And may be, integrate the code in the official Stockfish to improve its mate solving ability.
I'm afraid that finding the shortest possible mates wouldn't improve Stockfish's strength at all. Maybe if it was run in parallel on the GPU but then running something more relevant to the playing strength on the GPU would probably be better.
One thing that might be useful to carry over to Stockfish is to have an option to use a longer hash key in order to prevent incorrect results due to hash collisions in long analyses but that's for the developers to decide.
I wonder if somebody can give me some analysis with incorrect mate result by stockfish with a long analysis (hopefully with using one core so analysis is reproducable).
In the following position, SF and Crystal have trouble to find the mate in 13.
8/8/5N2/4K3/5N1p/1p3knR/1Pp2pr1/n1B2bbB w - - 8 2
that appears to be a lucky run, a second run took much longer - but no pv was posted as Iwas using the `go mate 13` command
info depth 92 seldepth 44 multipv 1 score mate 13 nodes 9968421809 nps 85038830 hashfull 1000 tbhits 0 time 117222 pv f6g8
bestmove f6g8 ponder f3g4
Ignacio wrote: ↑Mon Jun 21, 2021 4:52 pm
What about your mate solver?.
It's mostly ready but I had to attend to another project. I'll get back to it in a couple of weeks. Just need to test it in various chess GUIs pretty much.
Ignacio wrote: ↑Mon Jun 21, 2021 4:52 pm
What about your mate solver?.
It's mostly ready but I had to attend to another project. I'll get back to it in a couple of weeks. Just need to test it in various chess GUIs pretty much.