Unofficial stockfish 28.04.2013 with rtb tablebases support

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

Vinvin
Posts: 5290
Joined: Thu Mar 09, 2006 9:40 am
Full name: Vincent Lejeune

Re: Unofficial stockfish 28.04.2013 with rtb tablebases supp

Post by Vinvin »

j_romang wrote:Your download is way faster Josh ! :D
I'm also planning to share the 6 pieces tablebases (80Go), so maybe a torrent will be more usefull for that.
If the official compile use this TB and if it continues in the future, I'll download them.
zamar
Posts: 613
Joined: Sun Jan 18, 2009 7:03 am

Re: Unofficial stockfish 28.04.2013 with rtb tablebases supp

Post by zamar »

Thanks for doing the hard work....

Now it would be very interesting to see if we can get any ELO improvement with this...

I'd like to force loading all 5-men WDL bases directly into RAM and then see how it goes...

I'm just wondering if doing popcnt(pos.pieces()) every time in search is more expensive than keeping track of the total number of pieces under Position class...
Joona Kiiski
Vinvin
Posts: 5290
Joined: Thu Mar 09, 2006 9:40 am
Full name: Vincent Lejeune

Re: Unofficial stockfish 28.04.2013 with rtb tablebases supp

Post by Vinvin »

zamar wrote:...Now it would be very interesting to see if we can get any ELO improvement with this...
...
Note that tests with other engines show no measurable improvement with 6 pcs TB. The interesting part is the manual analysis with engine. It's painful to constantly show engine theorical draws (ex : R vs R+f&h pawns).
User avatar
geots
Posts: 4790
Joined: Sat Mar 11, 2006 12:42 am

Re: Unofficial stockfish 28.04.2013 with rtb tablebases supp

Post by geots »

velmarin wrote:Thanks for the engine.
I miss libstdc + +-6.dll, download by
https://code.google.com/p/wtfu/download ... l&can=2&q=

But breaks in my computer.

The Tablebases also can be downloaded from here,
Kindly Rises pot Joshua Shriver:
http://www.talkchess.com/forum/viewtopic.php?t=47758





Strictly as one good friend to another, who cares about you- when this all blows up 9 ways from Sunday- have you put much thought into an exit door.


Watch your back-

george
User avatar
velmarin
Posts: 1600
Joined: Mon Feb 21, 2011 9:48 am

Re: Unofficial stockfish 28.04.2013 with rtb tablebases supp

Post by velmarin »

geots wrote:
velmarin wrote:Thanks for the engine.
I miss libstdc + +-6.dll, download by
https://code.google.com/p/wtfu/download ... l&can=2&q=

But breaks in my computer.

The Tablebases also can be downloaded from here,
Kindly Rises pot Joshua Shriver:
http://www.talkchess.com/forum/viewtopic.php?t=47758
Strictly as one good friend to another, who cares about you- when this all blows up 9 ways from Sunday- have you put much thought into an exit door.
Watch your back-
george
I do not understand anything.

But thank you very much, George.
Milos
Posts: 4190
Joined: Wed Nov 25, 2009 1:47 am

Re: Unofficial stockfish 28.04.2013 with rtb tablebases supp

Post by Milos »

Vinvin wrote:
zamar wrote:...Now it would be very interesting to see if we can get any ELO improvement with this...
...
Note that tests with other engines show no measurable improvement with 6 pcs TB. The interesting part is the manual analysis with engine. It's painful to constantly show engine theorical draws (ex : R vs R+f&h pawns).
To have any measurable improvement, effective depth reached on average search where TBs are accessed should be at least 30-35.
This is impossible with TCs used for testing. Ergo, testing TB effect in the way ppl do engine testing (TCs with TPM of couple of hundreds of ms at best) is pointless.
Uri Blass
Posts: 10825
Joined: Thu Mar 09, 2006 12:37 am
Location: Tel-Aviv Israel

Re: Unofficial stockfish 28.04.2013 with rtb tablebases supp

Post by Uri Blass »

Milos wrote:
Vinvin wrote:
zamar wrote:...Now it would be very interesting to see if we can get any ELO improvement with this...
...
Note that tests with other engines show no measurable improvement with 6 pcs TB. The interesting part is the manual analysis with engine. It's painful to constantly show engine theorical draws (ex : R vs R+f&h pawns).
To have any measurable improvement, effective depth reached on average search where TBs are accessed should be at least 30-35.
This is impossible with TCs used for testing. Ergo, testing TB effect in the way ppl do engine testing (TCs with TPM of couple of hundreds of ms at best) is pointless.
Can you explain why?

I think that the main advantage of tablebases should be at fast time control when engines do many mistakes without them.

Uri
User avatar
velmarin
Posts: 1600
Joined: Mon Feb 21, 2011 9:48 am

Re: Unofficial stockfish 28.04.2013 with rtb tablebases supp

Post by velmarin »

I also was reluctant.
But the evidence is certain

Game of 1 minute (1 core) and stockfish and shows visits bitbases ..
See TBhits.
I guess that helps ...



Image
Milos
Posts: 4190
Joined: Wed Nov 25, 2009 1:47 am

Re: Unofficial stockfish 28.04.2013 with rtb tablebases supp

Post by Milos »

Uri Blass wrote:Can you explain why?

I think that the main advantage of tablebases should be at fast time control when engines do many mistakes without them.
You think totally wrongly.
Higher depth reached=>more TB access, more TB access=>better score.
Statistically it's quite obvious...
Could also look the other way around. Looking at TB at higher depth, means seeing a correct score earlier, meaning more chance to select a correct move.
This is also kind of obvious... ;)
j_romang
Posts: 79
Joined: Mon May 16, 2011 2:52 am

Re: Unofficial stockfish 28.04.2013 with rtb tablebases supp

Post by j_romang »

You are right Joona, the popcount seems a bad idea ; I'll try to keep track of the total number of pieces.
Currently there is a ~1% slowdown in regular search (when tablebases are not accessed).