BGonline.org Forums

Two concrete examples of using "View Statistics"

Posted By: Timothy Chow
Date: Wednesday, 20 January 2010, at 2:38 a.m.

In Response To: Two concrete examples of using "View Statistics" (Matt Cohn-Geier)

MCG wrote:

The math is kind of fun but could you figure this out OTB? I don't think I am that good.

Yes, this is the key question. My impression is that top players have developed some ability to estimate gammon wins and losses, basically by studying what the bots say and accumulating knowledge over time. I'm not at that level myself yet but from what I read on this forum I gather it's possible.

To do the kind of calculation I gave for the C-note position, what one has to learn is to estimate OTB how many of White's wins and losses are with the cube on 8. Very few people in the world (if any) can do that right now, but one reason is surely that the bots don't easily provide that information. If they did, then my guess is that the top players would gradually develop a feeling for those numbers.

Assuming that that is true, then the key point is that the math itself is no harder than the math you already do to make take decisions. You just estimate effective gammons and do the math as if they were real gammons, using the same gammon prices and raw take points. (O.K., strictly speaking this is true only when you don't have to worry about re-re-doubles; the math does get more complicated then, and perhaps too complicated to do OTB. But for take decisions in a 4-point match, or take decisions for redoubles in an 8-point match, the math is exactly the same as what one does today.)

You can turn off variance reduction and it will roll out much faster.

Ah...good tip! Thanks.

As mentioned this is not what I think of when I think of cube efficiency or recube vig.

The terminology might need to be changed to avoid this confusion.

Variance reduction is not a priority for me.

This is basically an issue of speed. I suspect that even with the trick you noted above, non-variance-reduced statistics will take a lot longer to produce than variance-reduced statistics with the same accuracy.

Post Response

Subject:
Message: