Re: [Bug-gnubg] Tutor mode

[Joern Thyssen]

On Mon, Aug 12, 2002 at 11:35:50AM +0100, Ian Shaw wrote
> Windows build 020804
> Tutor mode = Analysis settings =  Expert
> 
>     GNU Backgammon  Position ID: 29wBAByz3RkCAA
>                     Match ID   : QQmuAAAAAAAA
>     +24-23-22-21-20-19------18-17-16-15-14-13-+  O: GnuBg (Cube: 2)
>     | O  O  O     O  O |   |                  |  0 points
>     | O  O  O     O  O |   |                  |  
>     |             O  O |   |                  |  
>     |                  |   |                  |  
>     |                  |   |                  |  
>     |                  |BAR|                  |v 5 point match
>     |                  |   |                  |  
>     |                  |   |                  |  
>     |    O        X  X |   |                  |  
>     | X  O  X  X  X  X |   |    X             |  Rolled 43
>     | X  O  X  X  X  X |   |    X           X |  0 points
>     +-1--2--3--4--5--6-------7--8--9-10-11-12-+  X: ian
> 
>     1. Cubeful 0-ply    8/5 8/4                      Eq.:  +0.910
>        0.874 0.186 0.003 - 0.126 0.006 0.000
>         0-ply cubeful [expert]
>     2. Cubeful 0-ply    12/5                         Eq.:  +0.869 ( -0.041)
>        0.861 0.178 0.002 - 0.139 0.004 0.000
>         0-ply cubeful [expert]
> 
> 12/5 was marked doubtful so I asked for a hint. I did a World-Class++
> evaluation which reversed the rankings.
> 
> Double clicking on 12/5 to play the move still told me my move was doubtful
> and the analysis window only contained the 0-ply analysis.
> 
> I can see the logic behind this; I've asked it to Tutor on analysis setings
> and that is 0-ply. However, it doesn't play nicely. I can imagine it causing
> some confusion. 
> I think it would be better if the more thorough evaluation were pasted into
> the analysis, too, although I realise it isn't strictly "correct".

Tutor-mode will re-evaluate with the analysis settings, so the move is
still considered doubtful on 0-ply.

It's quite simple to let tutor-mode use the stored analysis from hint,
however, this introduces new problems. Remember, you cannot compare
evaluations on different plies. 

After being prompted with the doubtful move, you press hint. You now do
a 2-ply evaluation on move #10 and #11 -- this will move #10 and #11 up
to position #1 and #2 in the hint-dialog, because gnubg assumes 2-ply is
better than 0-ply even though the 2-ply moves have lower equity than the
0-ply moves. Now as you decide to stick with your move. tutor-mode can't
calcuate the equity difference because it would have to compare 2-ply
equities against 0-ply equities. A number of solutions exist:

(a) laissez faire -- live with the current implementation
(b) have gnu re-evaluate "interesting" moves on 2-ply -- interesting
    moves could be your selected move and any moves with higher equity.
(c) ignore the fact that the evaluations were done on different plies
    and calculate equity difference anyway.

So what should it be? (b) can be problematic if you did a 5-ply
evaluation or a rollout. It could take a while before gnubg finished
re-evaluating "interesting" moves....

(c) is equally bad to (a) as gnubg may think you move is quite good even
though it isn't.

The optimal solution is probably (b), but it's "expensive", both in
terms of the time it would take to implement it and in terms of the CPU
time spend re-evaluating the moves. Unless someone has a good idea I
suggest we stick with (a).

Jørn

-- 
Joern Thyssen, PhD
Vendsysselgade 3, 3., DK-9000 Aalborg, Denmark
+45 9813 2791 (private) / +45 2077 2689 (mobile) / +45 9633 7036 (work)