BGonline.org Forums

GNU-bg's cubeless eq. report in cube rollouts

Posted By: Chuck Bower
Date: Sunday, 8 February 2009, at 2:41 p.m.

I should post the GNU-bg bug bulletin board but first I think I'll ask the experienced bot-users here if what I contend makes sense.

If you run a rollout of a cube position in GNU-bg, at the top it shows a couple of values -- "Rollout cubeless equity", one for the particular matchscore and one for $-game. The way it apparently comes up with this number is simply adding the weighted individual outcomes (of simple wins, gammon wins,...) for the results when the cube is not immediately turned.

Here's the issue I have with this: GNU-bg (unlike SW) only rolls out a candidate until the game ends, and that includes when the game ends with a cubeturn. So in the no double branch the game often ends rather quickly (next roll or a few rolls down the line). Then GNU-bg evaluates what it thinks the gammon breakdowns are, posts them along with the above mentioned cubeless equities.

But there is another branch and that occurs when the cube is accepted. Since the person owning the cube is presumably well behind in the game (else the cubeturn was erroneous) it typically takes many more dicerolls for the bot to reach end-of-game. Its subsequent evaluation of the gammon breakdowns at that point is more reliable (because it goes deeper into the game) than for the other branch.

Of course the cube is a factor of two higher, but all you have to do is divide the cubeless equity value for the accepted-cube-branch by 2 to reach an estimate of the "rollout cubeless equity".

So, to me it would make considerably more sense to quote in the first line of the RO report this value based upon the cube-accepted branch as opposed to the current practice of using that value from the cube-unturned branch. Quite often these values disagree by a large amount. (See my RO on GammonU here: http://www.gammonu.com/members/board/config.cgi?read=167544 as an example.)

Does what I just said make sense or is there some faulty component in my argument?

Messages In This Thread

• GNU-bg's cubeless eq. report in cube rollouts
  Chuck Bower -- Sunday, 8 February 2009, at 2:41 p.m.
  • GNU-bg's cubeless eq. report in cube rollouts
    Rich Munitz -- Sunday, 8 February 2009, at 4:52 p.m.
    • GNU-bg's cubeless eq. report in cube rollouts
      Chuck Bower -- Sunday, 8 February 2009, at 10:05 p.m.
  • In match play it's even worse
    Maik Stiebler -- Monday, 9 February 2009, at 10:52 p.m.