Page 1 of 3
New MateFinder Release
Posted: Mon Dec 14, 2015 10:11 pm
by jhellis3
Updates:
1) Sync with latest SF.
2) Move to full hash key.
3) Increase max depth to 250.
4) Improve mate detection.
https://dl.dropboxusercontent.com/u/114 ... eFinder.7z
The archive includes the source and two very basic windows compiles.
Re: New MateFinder Release
Posted: Mon Dec 14, 2015 10:18 pm
by Dann Corbit
Thank you so much for your efforts.
I practically can't live without the latest matefinder stockfish.
It's the neatest thing since sliced bread.
It's really excellent on general tactical problems, not just checkmates.
Re: New MateFinder Release
Posted: Tue Dec 15, 2015 12:27 am
by Vinvin
As analyzing tools, would mind to add 2 patches constantly put back by official SF :
1) draw after 3 repetitions (and not 2) :
http://tests.stockfishchess.org/tests/v ... 227df1710b
2) bad corner bishop corner for pawn promotion.
Thanks !
Re: New MateFinder Release
Posted: Tue Dec 15, 2015 2:07 am
by jhellis3
1) I tried this, but it did not appear to actually solve the issue for me (at least in analysis), perhaps it works in actual gameplay?
2) Stockfish has this knowledge already, it is just quite specific, which makes it of limited efficacy. I can try to widen the coverage, but this is often quite hard to do if you are not willing to incorrectly score wins as draws (there is almost always another way to win), but many people are averse to the idea masking engine failings with false evals.
The problem is that if you miss even a single position which is a draw and don't score it as such, then you will not get the correct eval...
If you have the code ready to go though, I am more than happy to implement it.
Re: New MateFinder Release
Posted: Tue Dec 15, 2015 4:28 am
by MikeB
Compiling on the Mac:
Code: Select all
Step 2/4. Running benchmark for pgo-build ...
./matefinder bench 16 1 5000 default time > /dev/null
/bin/sh: line 1: 61046 Segmentation fault: 11 ./matefinder bench 16 1 5000 default time > /dev/null
make: *** [profile-build] Error 139
and just a normal (Non pgo) build
Code: Select all
Mac-Pro:src michaelbyrne$ ./matefinder
Segmentation fault: 11
Mac-Pro:src michaelbyrne$
Re: New MateFinder Release
Posted: Tue Dec 15, 2015 9:27 am
by stavros
how much elo weaker than normal sf?
Re: New MateFinder Release
Posted: Tue Dec 15, 2015 3:14 pm
by Jouni
Short test. Despite name it was not better mate solver than default SF. Also it was not better in tactics. But when I acticate "study" switch it solved some difficult studies impressively fast! I quess "study" is nullmove off?
Re: New MateFinder Release
Posted: Tue Dec 15, 2015 4:23 pm
by MikeB
Could not get it to run under Mac OS X 10.11.2 seg 11 error.
Re: New MateFinder Release
Posted: Tue Dec 15, 2015 6:31 pm
by jhellis3
Re: New MateFinder Release
Posted: Tue Dec 15, 2015 8:05 pm
by Dann Corbit
Jouni wrote:Short test. Despite name it was not better mate solver than default SF. Also it was not better in tactics. But when I acticate "study" switch it solved some difficult studies impressively fast! I quess "study" is nullmove off?
It solves many mates that SF cannot solve.
The easy mates, all the engines will solve them.
I do admit that I did not test the latest version yet.
But I have used it to solve many dozens of checkmates in a short time that the factory fresh SF could not solve in an hour.