Page 2 of 3
Re: Arasan test suite with minimal depth
Posted: Fri May 11, 2018 11:20 am
by op12no2
Dann Corbit wrote: ↑Thu May 10, 2018 8:53 pm
Grouped by 4, each position (of the 4) should have identical evaluation. If not, then there is an evaluation bug.
Found 2 bugs just now, trying that!
Re: Arasan test suite with minimal depth
Posted: Fri May 11, 2018 3:53 pm
by jdart
Micro-max has no hash table and no null move pruning.
I think with a hashtable, properly implemented PVS, and recursive null move, you should get >12 ply easily.
--Jon
Re: Arasan test suite with minimal depth
Posted: Fri May 11, 2018 4:02 pm
by xr_a_y
I think micromax has null move, hash table, ... but anyway, I get your point. Still looking for bugs
Re: Arasan test suite with minimal depth
Posted: Fri May 11, 2018 6:13 pm
by xr_a_y
xr_a_y wrote: ↑Fri May 11, 2018 6:57 am
Dann Corbit wrote: ↑Thu May 10, 2018 8:53 pm
Grouped by 4, each position (of the 4) should have identical evaluation. If not, then there is an evaluation bug.
Thanks I'll give it a shot today.
Cool, I found 3 little bugs in evaluation with this. Thanks a lot! now included in the unit test
Re: Arasan test suite with minimal depth
Posted: Fri May 11, 2018 8:00 pm
by Dann Corbit
Just writing a working chess program is quite an accomplishment.
I guess that you will have fun and progress quickly.
There are lots of experienced chess programmers who hang out in this forum and they can help you.
Re: Arasan test suite with minimal depth
Posted: Fri May 11, 2018 9:59 pm
by zenpawn
Dann Corbit wrote: ↑Thu May 10, 2018 8:53 pm
Grouped by 4, each position (of the 4) should have identical evaluation. If not, then there is an evaluation bug.
It looks like the castling rights aren't being adjusted on the color flips.
Re: Arasan test suite with minimal depth
Posted: Sat May 12, 2018 2:32 am
by Dann Corbit
zenpawn wrote: ↑Fri May 11, 2018 9:59 pm
Dann Corbit wrote: ↑Thu May 10, 2018 8:53 pm
Grouped by 4, each position (of the 4) should have identical evaluation. If not, then there is an evaluation bug.
It looks like the castling rights aren't being adjusted on the color flips.
Thanks for the bug report. I will look into it.
Re: Arasan test suite with minimal depth
Posted: Wed May 16, 2018 8:19 pm
by flok
Dann Corbit wrote: ↑Thu May 10, 2018 8:53 pm
Grouped by 4, each position (of the 4) should have identical evaluation. If not, then there is an evaluation bug.
But how do you handle PSQ? Because in
http://chessprogramming.wikispaces.com/ ... n+function the queen PSQ is not symmetrical.
Re: Arasan test suite with minimal depth
Posted: Wed May 16, 2018 8:38 pm
by Dann Corbit
In truth, all piece square tables should be non-symmetrical. I have some data used to make a piece square heat map, and not a single table is symmetrical.
So, if you have piece square tables coded that way, you can tell that they are the cause of the difference if you dump your eval term by term.
Re: Arasan test suite with minimal depth
Posted: Wed May 16, 2018 9:25 pm
by flok
Dann Corbit wrote: ↑Wed May 16, 2018 8:38 pm
In truth, all piece square tables should be non-symmetrical. I have some data used to make a piece square heat map, and not a single table is symmetrical.
So, if you have piece square tables coded that way, you can tell that they are the cause of the difference if you dump your eval term by term.
Ok. I was thinking that maybe the PSQ in the wiki was incorrect. Ok thanks.
But all other terms (like mobility) should be mirrorable, right?