|
From: | Robert-Jan Veldhuizen |
Subject: | Re: [Bug-gnubg] idea for faster rollout cube decision |
Date: | Mon, 04 Aug 2003 23:26:35 +0200 |
At 06:43 8/4/2003 +0000, you wrote:
On Mon, Aug 04, 2003 at 12:12:42AM +0200, Robert-Jan Veldhuizen wrote >> Wouldn't it be possible to just use the "old" DT equity (from an evaluation> or earlier rollout) and let this "half" rollout just update the ND equity? > And v.v. for rolling out DT only and keeping the old ND equity? This requires changes to the SGF file format, internal changes, and changes to the export, analysis, and GUI as everything assumes that the same setting is used for both ND and DT -- not an easy task as you may have guessed. Jørn
Okay. Well I still think it would be a very desirable option as it saves a lot of time but I don't expect anything soon ;)
I do think however that it would be more consistent if GNUBG treats the "cube decision" equities similar to move equites. Maybe the ND and DT equities could be displayed and structured just like if it were two candidate moves. Then the user could select either one or both and change rollout settings, just as with moves.
To give an example of a position where rolling out with GNUBG's current cube decision wastes a lot of time:
GNU Backgammon Position ID: OOewAVBsexFgIA Match ID : cAngAEAAAAAA +24-23-22-21-20-19------18-17-16-15-14-13-+ O: eberlein | X O O | O | X O O | 4 points | O O | | X O O | | O O | | O | | | | | | | | | | |BAR| |v 7 point match (Cube: 1) | | | | | X | | | | X | | | | X X X X | | O | On roll | O X X X X | | X X O | 0 points +-1--2--3--4--5--6-------7--8--9-10-11-12-+ X: ZorbaThe standard cube decision rollout will spend most of its time with O taking here, and then playing the game all to the end from there. All this is useless calculations really, since it is a HUGE drop (< -2 even). The only interesting equity here is if X doesn't double. Quite often X will double later, with a lot of cashes probably (truncating this part of the rollout, making it much quicker).
If I could just rollout the ND equity here, I estimate it would be 3 to 4 times faster, so definitely a big improvement if this could somehow be implemented I think.
--Robert-Jan (Zorba on FIBS)
[Prev in Thread] | Current Thread | [Next in Thread] |