TalkChess.com
Hosted by Your Move Chess & Games

Author Message
Jesús Muñoz

Joined: 13 Jul 2011
Posts: 692

Post subject: Doubt: perft estimate averaging N MonteCarlo samples.    Posted: Sun Jan 15, 2012 12:27 pm

Hello:

I have been thinking about an idea that will be surely clumsy: using 'perftmc' command of GNU 5.07.173b (which gives different values in each try), could I do a good estimate averaging results? Here is a typical output:

 Code: m=6.188431e+019 sd=5.191844e+015 ci(99%)=[6.187093e+019,6.189768e+019] n=2001154 971 sdn=2.322533e+020 t=37391.12s

It is an example of MonteCarlo Perft(14) for the initial position of standard chess. I thought about using NebiyuChess 1.42, which automatically interrupts MonteCarlo Perft at 5e+8 nodes more less... but always gives the same results in different tries, so I discarded it. This is the reason why I thought about GNU.

My question is the following: if I run N MonteCarlo samples of the same position and I stop all of them at the same point (with the same number of nodes ± a very little amount), can I average the results in this way?

 Code: m_1, m_2, ..., m_N (Average m) = = (1/N)·(m_1 + m_2 + ... + m_N) sd_1, sd_2, ... sd_N (Average standard deviation) = = sqrt{(1/N)·[(sd_1)² + (sd_2)² + ... + (sd_N)²]}

Or, taking into account that different samples have different number of nodes:

 Code: = [(n_1)·(m_1) + (n_2)·(m_2) + ... + (n_N)·(m_N)]/(n_1 + n_2 + ... n_N) = sqrt{[(sdn_1)² + (sdn_2)² + ... + (sdn_N)²]/(n_1 + n_2 + ... n_N)}

Where sdn_i = (sd_i)·sqrt(n_i) and GNU prints these numbers. I think that n_i ~ 1e+8 or 5e+8 nodes in each sample is accurate enough, but comments are welcome.

Are my assumptions correct or I have failure concepts? In the case that I am right (very unlikely), the estimate would be good/accurate or is it a waste of time because of the size of the standard deviation (or other issues)? Thanks in advance.

Regards from Spain.

Ajedrecita.
_________________
Six Fortran 95 tools.

Chess will never be solved.
 Display posts from previous: All Posts1 Day7 Days2 Weeks1 Month3 Months6 Months1 Year Oldest FirstNewest First
Subject Author Date/Time
Doubt: perft estimate averaging N MonteCarlo samples. Jesús Muñoz Sun Jan 15, 2012 12:27 pm
Daniel Shawul Sun Jan 15, 2012 1:31 pm
Jesús Muñoz Mon Jan 16, 2012 8:35 pm
Jesús Muñoz Fri Jan 20, 2012 3:27 pm
Jesús Muñoz Sun Jan 22, 2012 4:09 pm
Jesús Muñoz Thu Jan 26, 2012 4:44 pm
Jesús Muñoz Fri Feb 17, 2012 8:00 pm
Jesús Muñoz Fri Mar 02, 2012 5:26 pm
Jesús Muñoz Fri Mar 09, 2012 3:50 pm
Jesús Muñoz Fri Mar 16, 2012 4:57 pm
Jesús Muñoz Fri Mar 23, 2012 3:22 pm
Jesús Muñoz Fri Mar 30, 2012 3:11 pm
Jesús Muñoz Tue Apr 10, 2012 7:21 pm
Peter Österlund Tue Apr 10, 2012 9:24 pm

 Jump to: Select a forum Computer Chess Club Forums----------------Computer Chess Club: General TopicsComputer Chess Club: Tournaments and MatchesComputer Chess Club: Programming and Technical DiscussionsComputer Chess Club: Engine Origins Other Forums----------------Chess Thinkers ForumForum Help and Suggestions
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum