Page 1 of 1
Problem with KRRPKQ
Posted: Wed Aug 09, 2006 10:32 am
by Kirill Kryukov
Hi all! I was doing some analysis using Wilhelm 1.49 when I encountered that it does not show the full line for some positions. I tracked it down to this position (white to move):
Code: Select all
8/8/4q3/5R2/8/5R2/1KP3k1/8 w - - 0 1
Wilhelm sais white will win in 100 moves, but can't show the line, apparently because one position can't be probed.
The position which can't be probed is one move after (black to move):
Code: Select all
8/8/4q3/5R2/8/8/1KP2Rk1/8 b - - 0 1
In this position Wilhelm sais "Unknown", it does not know if it is winning or not.
I believe my KRRPKQ tablebase set is full and correct - 16 files with MD5 (I just re-checked the MD5):
Code: Select all
91b0a26b66679ac67263f27da908ecf3 krrpkq.0.nbb.emd
13bfd39728547698f32f7f74880bfe8f krrpkq.0.nbw.emd
a1f9f7051b3e8649dbcdfaa3de749748 krrpkq.1.nbb.emd
bdddde20f42e4314b4afe38e53750f21 krrpkq.1.nbw.emd
cae48bd7c39558a7d631dd0b2fefe034 krrpkq.2.nbb.emd
4a9f407cb99e444f1ff2b8b62b14421b krrpkq.2.nbw.emd
f4737ce74a984fc862924b87a2543a76 krrpkq.3.nbb.emd
f41f24918cfe90d5e5062ba706bcf65a krrpkq.3.nbw.emd
329f7559d0d517bb61806444e7aa6e0a krrpkq.4.nbb.emd
42cc7eaeb7a2846dd45b61257573c22f krrpkq.4.nbw.emd
237bc1d628b44a38461248c8b069134a krrpkq.5.nbb.emd
47726d10ccf7d59b56369ce4fc6e1705 krrpkq.5.nbw.emd
3e5e7c179ba3bf904d91a2deed54cd7e krrpkq.6.nbb.emd
7d72013a837fd6d9ae1e16cea62b7905 krrpkq.6.nbw.emd
b677018e5f68eea90389f3e8bf55a7a1 krrpkq.7.nbb.emd
d9579a77f95890318a7a6ffc08abe5e4 krrpkq.7.nbw.emd
Can anyone try that position? I worrying if the problem is in Wilhelm or in the tablebase files. Any help will be very much appreciated!
Posted: Wed Aug 09, 2006 9:59 pm
by kalyanasundaram
Hello
The problem seems to come from wilhelm. I tested the position with wilhelm and have the same results as Kirill
But all works well with shredder classic and fritz 9 (mate in 127 after Rf2+ )
Regards
kalya
Posted: Fri Aug 11, 2006 8:23 am
by Kirill Kryukov
Thanks for checking! I am relieved to know that the problem is with Wilhelm and not with the tablebase. I emailed to Rafael (Wilhelm author) about it.
Posted: Thu Aug 17, 2006 9:02 am
by Kirill Kryukov
One more position where Wilhelm has problem, this time in KQBKQP:
4b3/8/8/8/6Q1/k1K1P3/5q2/8 w - -
(white to move)
The position is supposed to be win in 128 for white, but Wilhelm does not show the line, because it can't probe position which occurs after 1.Qb4+
Code: Select all
4b3/8/8/8/1Q6/k1K1P3/5q2/8 b - - 0 1
(black to move)
In this position Wilhelm sais "Unknown", though the tablebase is installed, compelte and verified.
Blind guess ...
Posted: Thu Aug 17, 2006 11:11 am
by guyhaw
Blind guess: I wonder if this is another case of the oft-forgotten 8-bit/16-bit parameter raising its head. Anyway, Rafael's 'around' so will presumably pick up on yr queries soon.
g
Re: Blind guess ...
Posted: Thu Aug 17, 2006 1:37 pm
by Kirill Kryukov
guyhaw wrote:Blind guess: I wonder if this is another case of the oft-forgotten 8-bit/16-bit parameter raising its head. Anyway, Rafael's 'around' so will presumably pick up on yr queries soon.
g
Hi Guy! Rafael replied to my email and said he will look into the problem. He has to download the relevant tablebase first, which will take some time.
I think it is not 8-bit/16-bit issue, because if it was so then Wilhelm would presumably now work at all with a given tablebase, but instead it seems working fine for most position, and fails for just very few positions.
Posted: Fri Aug 18, 2006 5:23 am
by Kirill Kryukov
I found same problem with KQNKQP tablebase too. I'll post position if needed.
Posted: Sat Aug 19, 2006 11:23 am
by Rafael B. Andrist
I found same problem with KQNKQP tablebase too. I'll post position if needed.
Let me guess... the crucial position is a mate in 127 moves? I think I have found the problem and know how to fix it.
regards
Rafael Andrist
Posted: Sat Aug 19, 2006 12:00 pm
by Kirill Kryukov
Rafael B. Andrist wrote:Let me guess... the crucial position is a mate in 127 moves? I think I have found the problem and know how to fix it.
regards
Rafael Andrist
Hi Rafael, yes, the position is mate in 127 moves. I'm glad you are on the track of the problem! Did you successfully download any of the problematic tablebase sets? Looking forward to the fixed version! Thanks for your work!
Posted: Sat Aug 19, 2006 12:20 pm
by Wulff
Hi Andre!
Will there be a new version of Wilhelm available ??
