| |
BGonline.org Forums
Cube error rates
Posted By: Robert-Jan Veldhuizen (Zorba) In Response To: Cube error rates (Frank Berger)
Date: Sunday, 26 July 2009, at 1:10 p.m.
This is not about lookahead (other than that it might be more accurate, as the bots' NNs are imperfect). The ND equity is NOT the never double equity, it's the equity of not doubling NOW.
Janowski centered cube equities are supposed to take future cube usage into account (from both players). This is the whole point of Janowski. It's supposed to give a "live cube" value.
Never doubling by the player (but possibly by the opponent) can be calculated with Janowski too, this is "opponent owns 1-cube" and is a potentially very different value.
Never doubling by either side is simply cubeless equity (or dead cube, if you wish), of course.
The example I gave is pretty typical, there might be a bear-off cube decision that comes pretty close I guess, or some simple bear-in versus an anchor. A position might be worth (on 0-ply!)
75% wins, 25% losses 0.5 cubeless (never a double by either side) 0.7 No Double (not doubling now, but possibly by either player in the future) 0.9 Double/Take (it's a take because of recube vig)
0.45 opponent owns 1-cube, i.e. NEVER DOUBLING by the player, but possibly by opponent. For money, no Jacoby, this is simply D/T divided by 2. It's lower than cubeless, because opponent has cube vig.
Now, 0.7 is way different than 0.45 or 0.5. The 0.7 is much higher as it takes future cube usage into account, by applying Janowski for centered cube.
If you have a strong cube now, you'll often have a cube later too if you don't double now. Janowski takes this into account. Even at 0-ply, the ND equity takes future cube usage into account. This is why the figure is so high!
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.