perft(15)

Discussion of chess software programming and technical issues.

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Dann Corbit
Posts: 9981
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: perft(15)

Post by Dann Corbit » Sat Aug 26, 2017 6:58 am

ankan wrote:
Dann Corbit wrote: Do you have stored statistics, such as how many mates are in the first 15 ply, how many stalemates, etc.?
Unfortunately no. I can modify my program to count these statistics - but it will probably become quite a bit slow. I would guess counting these will for first 14 ply would take about the same time as the perft 15 computation took.
I strongly suspect that your move generator (since it must detect mates to get the correct answer ) would create the strongest mate prover eve imagined.

Isn't it time to give Chest319 some competition?
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

ankan
Posts: 71
Joined: Sun Apr 21, 2013 1:29 pm
Full name: Ankan Banerjee
Contact:

Re: perft(15)

Post by ankan » Sat Aug 26, 2017 7:10 am

Dann Corbit wrote: I strongly suspect that your move generator (since it must detect mates to get the correct answer ) would create the strongest mate prover eve imagined.

Isn't it time to give Chest319 some competition?
Yes, that's the next thing I plan to attempt. However the main issue is alpha-beta pruning (and current parallel search algorithms based on that) is not something that scales well to thousands to millions of threads. I have some plans and am going to try out a few things. Will update if I have any success with it.

Dann Corbit
Posts: 9981
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: perft(15)

Post by Dann Corbit » Sat Aug 26, 2017 7:31 am

ankan wrote:
Dann Corbit wrote: I strongly suspect that your move generator (since it must detect mates to get the correct answer ) would create the strongest mate prover eve imagined.

Isn't it time to give Chest319 some competition?
Yes, that's the next thing I plan to attempt. However the main issue is alpha-beta pruning (and current parallel search algorithms based on that) is not something that scales well to thousands to millions of threads. I have some plans and am going to try out a few things. Will update if I have any success with it.
I am so excited that someone with your immense talent is looking at this.

I expect the best mate finder (by at least two orders of magnitude) will be the result.

You have to really understand these GPU boards to get the horsepower out of them. I am really geeked to see people of great talent putting their brains at risk to solve these difficult problems.
:-)!
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

User avatar
cdani
Posts: 2104
Joined: Sat Jan 18, 2014 9:24 am
Location: Andorra
Contact:

Re: perft(15)

Post by cdani » Sat Aug 26, 2017 10:20 am

Nice work! Congratulations!!

ankan
Posts: 71
Joined: Sun Apr 21, 2013 1:29 pm
Full name: Ankan Banerjee
Contact:

Re: perft(15)

Post by ankan » Sat Aug 26, 2017 12:28 pm

Thanks all. Wouldn't have been possible without encouragement from everyone in this forum (especially Steven).

smatovic
Posts: 822
Joined: Wed Mar 10, 2010 9:18 pm
Location: Hamburg, Germany
Full name: Srdja Matovic
Contact:

Re: Perft(15): comparison of estimates with Ankan's result.

Post by smatovic » Sat Aug 26, 2017 4:39 pm

Kudos from Germany......years of compute cycles compressed into 5 days, impressive.

--
Srdja

User avatar
Ajedrecista
Posts: 1397
Joined: Wed Jul 13, 2011 7:04 pm
Location: Madrid, Spain.
Contact:

Re: Perft(15).

Post by Ajedrecista » Sat Aug 26, 2017 6:34 pm

Hello Dann:
Dann Corbit wrote:Do you have stored statistics, such as how many mates are in the first 15 ply, how many stalemates, etc.?
The current shortest known stalemates are 19-ply stalemates. It was discussed in the following thread:

estimating the number of possible stalemates in perft(n)

Regarding checks, checkmates, discovered checks and so on, you can get interesting info for some depths at this site:

Statistics on chess games
Dann Corbit wrote:I strongly suspect that your move generator (since it must detect mates to get the correct answer ) would create the strongest mate prover eve imagined.

Isn't it time to give Chest319 some competition?
It would be great. By the way, should not ChestUCI 5.2 be somewhat better than Chest 319? Just asking.

Regards from Spain.

Ajedrecista.

bhamadicharef
Posts: 31
Joined: Fri Nov 25, 2016 9:14 am
Location: Singapore
Contact:

Re: perft(15)

Post by bhamadicharef » Thu Aug 31, 2017 6:46 am

What is Chest319 ? What method does it use ?
Brahim HAMADICHAREF
Singapore

bhamadicharef
Posts: 31
Joined: Fri Nov 25, 2016 9:14 am
Location: Singapore
Contact:

Re: Perft(15): comparison of estimates with Ankan's result.

Post by bhamadicharef » Thu Aug 31, 2017 6:50 am

Nice job Ankan ! With 4 DGX-1 the job is getting faster then :-)
I look forward to try your tool on my servers at work too !
Brahim HAMADICHAREF
Singapore

Dann Corbit
Posts: 9981
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: perft(15)

Post by Dann Corbit » Thu Aug 31, 2017 6:50 am

bhamadicharef wrote:What is Chest319 ? What method does it use ?
Chest is the best available mate solver, by Heiner Marxen.
It comes in two formats:
Command line driven and also as a UCI engine. The UCI version has lots of options.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

Post Reply