Réf. : [Bug-gnubg] Current categorization of dou bling mistakes (new thread)

[Massimiliano . Maini]

>I still strongly disagree with the new categorization of doubling mistakes.
>
>Here's a doubling decision I had tonight on GG against Martin Krainer in a 7
>pointer:
>Position ID : m2zgAUTLNiAWAw
>Match ID : UQngAAAAAAAA
>
>Cube analysis
>2-ply cubeless equity  +0,566 (Money:  +0,538)
>   64,3%  34,9%   1,1% -  35,7%  10,7%   0,2%
>Cubeful equities:
>1. Double, take         +0,831
>2. Double, pass         +1,000  ( +0,169)
>3. No double            +0,743  ( -0,088)
>Proper cube action: Redouble, take
>
>I failed to redouble here, underestimating my gammon wins. Incidentally,
>with Position ID, you will realize TG is not even an issue to consider.

The double categorization has been changed a few hours before your mail
(see mail from Holger). Now it has :

      Wrong double  (below DP)
      Wrong double  (above TG)
      Missed double (below CP)
      Missed double (above CP)

With your position (DP=46.3%, CP=68.4%, W%=64.3%) the missed double is
now seen as a "missed double below CP".

>But Gnubg classified this mistake as 'Missed double around CP' : I'm sorry I
>have to say this doesn't make sense *at all* to me. *No way* this position
>can be TG, still that is what the wording implies, as Gnubg basically tells
>me I forgot to cash, eventhough it's a huge take (I *do* know it doesn't
>tell me that, but it sure sounds like it, the use of the words Cash Point is
>totally counter-intuitive).

No : GNUbg was telling you that you forgot to double, and that you were
closer to CP than to DP. It was a take since you were below CP, like the
(very) new categorization stresses more. TG has nothing to do here.
The (very) new categorization just tells you that you missed a double
between DP and CP (below CP) or between CP and TG (above CP) : it no longer
tries to tell that you were closer to CP or to DP (or to TG).

>Now I realized where the pb comes from and why Gnubg now classifies a vast
>majority of missed doubles as 'Missed double around CP' : you can sum it up
>in just one word, and that is EFFICIENCY. In BG, you want to send over cubes
>as *efficient* as possible, in other words, as close to your CP as possible.

I disagree : if possible, you should double as soon as you are above DP
(and below TG). In the continuous BG model behind dead/live cube models
this always happens (you never jump from below DP to above TG, in fact,
you never jump, that's why it is continuous). If you play on above DP hoping
to become TG, you're playing suboptimally (or you're gambling, or you have to
go in 5 minutes and hoping for a quick end).

Maybe I'm just missing something but ....

>The above-posted doubling categorization therefore illustrates the following
>problem : a majority of cubes will be sent with some degree of efficiency
>(varying with the level of players), which very often translates in maths
>terms into a double that is closer to CP than DP. Only exceptions occurr
>when volatility is *very* high on the next roll/next exchange, or in a last
>roll position (this is actually a very specific sub-case of the first case)
>or in long matches when you have a huge lead and your market window is very
>tiny.
>
>In the position referred to, my Doubling Point was 46.3% and my CP 68.4% :
>have a guess, in 90% of cases, which one I will be closer to when I
>redouble...?

You should have reduobled as soon has your W% was above 46.3% : that's the
optimal play accoriding to the underlying model.

>So, because of efficient doubles, we end up with a predominant 'Missed
>double around CP' category ; this introduces confusion because it now
>intuitively sounds as you missed a double/pass (because of the words Cash
>Point) eventhough this category still refers to a majority of Missed
>double/take!!

The (very) new categorization fixes this (using below/above CP).
But I still don't grasp the efficiency stuff (I'm not saying it's wrong ...).

>Incidentally, Gnubg also went wrong a couple of rolls later in that very
>same game, when I decided to redouble :
>
>Position ID : 29qAAQybbQwZAA
>Match ID : UQngAAAAAAAA
>
>Cube analysis
>2-ply cubeless equity  +0,447 (Money:  +0,441)
>   69,6%   9,2%   0,1% -  30,4%   4,4%   0,0%
>Cubeful equities:
>1. No double            +0,752
>2. Double, pass         +1,000  ( +0,248)
>3. Double, take         +0,681  ( -0,071)
>Proper cube action: No redouble, take (22,4%)
>
>Gnubg classified this as 'Wrong double around CP' eventhough as you can see,
>it's a huge take, so cashing is not even an issue unless the opponent makes
>a triple whopper (a 0.319 blunder).

[You meant "missed" and not "wrong", right ?]
Same as bove, the (very) new categorization would give "missed double below CP"
(aka Double/Take).


MaX.