Tuning Futility Margin

Discussion of chess software programming and technical issues.

Moderators: hgm, Dann Corbit, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
tomitank
Posts: 235
Joined: Sat Mar 04, 2017 11:24 am
Location: Hungary

Tuning Futility Margin

Post by tomitank » Mon Jun 19, 2017 9:08 am

Hi!

I've read that it can be tuned with fixed-depth search.
Minimum ELO loss should be sought.
I'm testing this at depth 8. (Self game)
(I tried at depth <= 4)
My average depth is 10-14.
Does anyone have a test result of the "recommended" ELO loss?

Thx, Tamas

User avatar
Evert
Posts: 2929
Joined: Fri Jan 21, 2011 11:42 pm
Location: NL
Contact:

Re: Tuning Futility Margin

Post by Evert » Mon Jun 19, 2017 10:20 am

This doesn't make much sense to me. Surely the "minimal Elo-loss" would be 0 (no futility)?

Moreover, I can imagine that the optimal value will depend on time control. I would just tune using your target time control, and maybe verify that it is not a regression at other time controls.

tomitank
Posts: 235
Joined: Sat Mar 04, 2017 11:24 am
Location: Hungary

Re: Tuning Futility Margin

Post by tomitank » Mon Jun 19, 2017 11:04 am

While the ELO loss around zero, then i winning extra speed -> same time, more depth -> extra ELO .... or?

Idea from: http://people.csail.mit.edu/heinz/dt/node18.html

Ernst A. Heinz:
Although extended futility pruning is theoretically unsound, extensive experiments with our master-strength chess program DARKTHOUGHT show that it works markedly well in practice. Even at fixed search depths, extended futility pruning exhibits hardly any loss of tactical strength while shrinking the search trees by 10%-20% on average as compared with normal futility pruning. Furthermore, extended futility pruning combines nicely with a conservatively limited variation of razoring that reduces the search trees at fixed search depths > 10 plies by additional 5%-15% on average. Last but not least, the presented pruning schemes scale very well with search depth and reap ever more benefits at higher depths.

jdart
Posts: 4070
Joined: Fri Mar 10, 2006 4:23 am
Location: http://www.arasanchess.org

Re: Tuning Futility Margin

Post by jdart » Mon Jun 19, 2017 1:21 pm

When he says "hardly any loss of tactical strength," he does not mean playing strength goes down overall.

The whole point of tuning is to raise Elo so you have to tune for improvement over enough games to have confidence (SPRT test for example).

--Jon

User avatar
Evert
Posts: 2929
Joined: Fri Jan 21, 2011 11:42 pm
Location: NL
Contact:

Re: Tuning Futility Margin

Post by Evert » Mon Jun 19, 2017 2:41 pm

tomitank wrote:While the ELO loss around zero, then i winning extra speed -> same time, more depth -> extra ELO .... or?
Maybe. I don't think it follows logically, though.

Either way: your question is poorly formulated. You ask how to tune futility for "minimal Elo loss at fixed depth". The minimum Elo loss at fixed depth is for a futility margin of infinity, so no futility pruning.* This is not what you intend.

*Assuming that futility pruning actually is an Elo loss at fixed depth.

Also, Elo, not ELO. It's a proper name, not an abbreviation.

tomitank
Posts: 235
Joined: Sat Mar 04, 2017 11:24 am
Location: Hungary

Re: Tuning Futility Margin

Post by tomitank » Mon Jun 19, 2017 2:42 pm

What time management do you recommend?
I have i3-6100, and my average depth is 10-14 / 1 sec.
10s + 0.1 is correct?

Tamas

tomitank
Posts: 235
Joined: Sat Mar 04, 2017 11:24 am
Location: Hungary

Re: Tuning Futility Margin

Post by tomitank » Mon Jun 19, 2017 3:28 pm

Nemzetközileg ismert neve Elo (gyakran nagybet&#369;kkel ELO, bár nem bet&#369;szó). Nevét Él&#337; Árpád (angol, külföldön ismertebb nevén Arpad Elo) magyar születés&#369; amerikai fizikaprofesszorról kapta.
--------------------------------------------------------------------------------
Internationally known as Elo (often in capital letters ELO, though not an acronym). His name was Árpád Él&#337; (English, more commonly known as Árpád Elo) from the Hungarian-born American physics professor.

jdart
Posts: 4070
Joined: Fri Mar 10, 2006 4:23 am
Location: http://www.arasanchess.org

Re: Tuning Futility Margin

Post by jdart » Mon Jun 19, 2017 7:05 pm

10s + 0.1 is (mostly) what Stockfish uses.

--Jon

Post Reply