Does some of you change (or even discard) PST values when in FRC (or other variants) ?
Any other eval feature you tuned for FRC ?
PST for FRC
Moderators: hgm, Rebel, chrisw
-
- Posts: 1871
- Joined: Sat Nov 25, 2017 2:28 pm
- Location: France
-
- Posts: 584
- Joined: Fri Mar 30, 2018 7:20 am
- Full name: Andreas Matthies
Re: PST for FRC
I haven't changed anything in the FRC evaluation for now. And the result is really bad: https://github.com/Matthies/RubiChess/issues/203
Would be interesting if anybody with more FRC experience would answer here.
Andreas
-
- Posts: 570
- Joined: Mon Jul 20, 2015 5:06 pm
-
- Posts: 178
- Joined: Wed Nov 13, 2019 1:36 am
- Full name: Jonathan Kreuzer
Re: PST for FRC
SlowChess uses the same evaluation for regular chess and FRC. Compared to some similar strength programs it does statistically better in FRC than standard, although against Stockfish 11 FRC didn't help and scored about the same (more testing would be needed to say for sure.)
I'm not sure why it would do better in FRC. Possible positives for FRC are:
-Horizontal symmetry for the PST (and anything else that can be).
-Handling for King-Side related stuff and opposite castling is all through generalized terms, varied castling sides or not castling happens more in FRC.
-The automatic eval tuning includes training games from FRC (just like 5% of games). It seemed to help FRC and be even or slightly positive for standard chess, but didn't test enough to be sure. I wonder if just doing FRC games would work to mix up piece positions better so generalized terms matter more.
-Maybe opening play is strong now, but that seems unlikely to me since it usually doesn't feel that way. Though I have been using 2-moves book to test changes for 2.0 and 2.1
I'm not sure why it would do better in FRC. Possible positives for FRC are:
-Horizontal symmetry for the PST (and anything else that can be).
-Handling for King-Side related stuff and opposite castling is all through generalized terms, varied castling sides or not castling happens more in FRC.
-The automatic eval tuning includes training games from FRC (just like 5% of games). It seemed to help FRC and be even or slightly positive for standard chess, but didn't test enough to be sure. I wonder if just doing FRC games would work to mix up piece positions better so generalized terms matter more.
-Maybe opening play is strong now, but that seems unlikely to me since it usually doesn't feel that way. Though I have been using 2-moves book to test changes for 2.0 and 2.1
-
- Posts: 893
- Joined: Mon Jan 15, 2007 11:23 am
- Location: Warsza
Re: PST for FRC
This is interesting. A friend added FRC support to Rodent and after playing a few games I know that its eval is sub-optimal for this chess variant. My engine seems to be reluctant to take central space in FRC, even though in normal games it does so willingly. The reason is that it does not have a big penalty for a pawn staying on d2/e2. This is intentional - together with a fianchetto bonus it was meant to make it accept hypermodern openings more readily, and rather high bishop mobility does the rest (without book, I still get 1.e4 as the first move). FRC may require changing my piece/square tables.
Pawel Koziol
http://www.pkoziol.cal24.pl/rodent/rodent.htm
http://www.pkoziol.cal24.pl/rodent/rodent.htm