Cube 119 - Ahead 2-0 3-away

[sixty_something]

an excellent reply, Zorba .. while i haven't had the time to read and attempt to "penetrate" Doug Zare's comments, what you have summarized as a Too good to cube/Take situation makes really good sense, to me

this thread within a thread has been most interesting .. the problem itself, was solved by most answering the initial poll and appears to almost be a trivial one in comparison to the issues raised by these responses

when i have the chance, i will split off these replies and retitle the new thread as Too good to cube/Take .. it is just possible this will come up again and perhaps even be found in analysis in few more positions .. for now, continue to reply here, if desired, on this Too good to cube/Take issue .. it will likely be a few days before i can get around to moving it to a new thread

[dorbel]

Thank you Zorba, a much clearer explanation than raccoon's. However, this sort of position is normally described by humans and bots as being no double/take, even though the equity in emg is over 1 before doubling. It is just a question of semantics. The position as originally posted though, has gnu clearly saying "Too good to double/take". Gnu on my machine says "no double/take" for the identical position and settings. Any clues on that?
IMO the traditional description is better.

[Zorba]

There are indeed many players and some bots that will label "too good to double, take" as "no double, take" or worse, "not good enough, take". Maybe a matter of taste or habit, but I think it might also just be an oversight or lack of understanding sometimes.

Of course, the correct cube action for both players is the same: do not double; and after a wrong double, a take for the opponent.

However, there is still an important difference between "too good, take" and "no double, take", and that is, what happens when opponent makes a mistake and passes? The chance of an opponent making a mistake after you double is often some extra equity you get from doubling, and can make a theoretically wrong double correct in practice.

If you get an incorrect pass when it's actually "no double (not good enough), take", you are probably gaining a huge chunk of equity. A typical GNUBG output might read:


+0.600 No Double
+0.550 Double, Take
+1.000 Double, Pass

Correct cube action: No Double, Take (12.5%)


That figure in brackets, 12.5% tells you when it becomes profitable in practice, to double anyhow: If the chance of opponent incorrectly passing is 12.5% or more, doubling is correct! The reason is that you gain 1.000-0.600=0.400 on an incorrrect pass, and the loss comes at only 0.550-0.600= -0.050.



Now, by contrast, if you get an incorrect pass when it's actually "too good to double, take", you are still losing equity! After all, a pass gains you exactly 1 point, but you were too good, so playing on would have gained you more than 1 point.

So, it is NEVER correct to double a "too good, take", not even if your opponent would somehow blunder with a pass.

Here's an example of "too good, take", X launched a rather succesful blitz, two of O's checkers are on the bar, the 6-5-4 points are made, but the position is still far from "gin" for X:

    GNU Backgammon  Position ID: 4NvgAWBqd4gBJA
                    Match ID   : cAmlABAAGAAA
    +24-23-22-21-20-19------18-17-16-15-14-13-+  O: samirah
    | X     X        O | O | O  O           X |  1 point
    |                O | O | O  O           X | 
    |                O |   |                  | 
    |                O |   |                  | 
    |                O |   |                  | 
    |                  |BAR|                  |v 5 point match (Cube: 1)
    |                  |   |                  | 
    |                  |   |                O | 
    |             X  X |   |                O | 
    |          X  X  X |   |                O |  Rolled 21
    |    X  X  X  X  X |   |          X     O |  3 points
    +-1--2--3--4--5--6-------7--8--9-10-11-12-+  X: You

X on roll (clockwise), cube action?

Cube analysis
2-ply cubeless equity  +1,475 (Money:  +1,107)
   78,5%  56,7%   1,6% -  21,5%   4,4%   0,2%
Cubeful equities:
1. No double            +1,089
2. Double, pass         +1,000  ( -0,089)
3. Double, take         +0,698  ( -0,391)
Proper cube action: Too good to double, take


If you incorrectly double here, and opponent passes, then your opponent is making a huge blunder: he loses 0.302 equity compared to a correct take.

However, even after that huge blunder, the doubler still loses 0.089 equity over simply playing on.

It's just that the doubler would have lost much more, namely 0.391 equity, if opponent had correctly taken his blunderish cube. But it's still a loss.


(N.B.: My original example with numbers wasn't good, more realistic is something like:
80% total wins, 60% gammons (i.e. a 75% gammon rate), 20% losses
to cause a "too good, take" at 2-away, 4-away)

(N.B. 2: the position in the rec.games.backgammon thread also comes out as a "no double, take" on my GNUbg, probably has to do with different match equity tables being used)

[playBunny]

However, this sort of position is normally described by humans and bots as being no double/take, even though the equity in emg is over 1 before doubling. It is just a question of semantics.

Humans, maybe, but GnuBg wouldn't know what semantics were if you spent 10 years trying to tell it. It does know about comparing numbers and making branches in program flow. To a bot it's a mathematical decision, not a semantic one. The fact that one GnuBg evaluation says Too good to double/take and the other says No double/take demonstrates that.

The position as originally posted though, has gnu clearly saying "Too good to double/take". Gnu on my machine says "no double/take" for the identical position and settings. Any clues on that?

The settings are not identical. You're using the g11 MET. Try the position with Woolsey's.