Re: [Bug-gnubg] Tutor mode: does not give notification about doubles (sample match+ instructions included)

[Jim Segrave]

On Tue 25 Feb 2003 (13:30 +0100), Scott Steiner wrote:
> Hi,
> 
> I posted this error some time ago but I did not receive any reply.  I am
> posting it again, this time with a sample match and instructions on how
> to reproduce the error.
> 
> I usually have my matches analysed at 2ply and go through them looking
> at my errors.  When reviewing them after the analysis, I usually turn on
> tutor mode and play on from an error with the correct move.  I have
> noticed that as I play on and I eventually reach a position where I
> should double, I don't get the usual popup window telling me if I want
> to rethink or play anyway.  However, when I go back to that position
> where the double should have come and I use the hint function, then
> gnubg tells me that that was a double, so it seems that there is
> something wrong with the tutor mode on doubling in this case.  Checker
> play is fine.  I have tutor mode enabled for both checker play and cube
> decisions.
> 
> Included is a sample match which will display the error in a slightly
> modified form - I will enable tutor mode and purposely not double where
> I should have doubled and you will see that I get no notification. 
> Proceed as follows please:
> 
> - open the analysed sample match in gnubg
> - open the game record window and the annotation window
> - in the game record window select game 3
> - click on move 6 for user and you will see in the annotation window
> that this was a correct double
> - go to the settings menu and enable tutor mode for both cube and
> chequer play in the options
> - now click on the middle of the board as usual to roll dice without
> doubling before
> - you should now notice that gnubg thinks for a few moments before it
> displays the dice and it eventually rolls the dice and displays them
> without giving a notification about the wrong cube decision
> 
> I am using build Feb 15 2003 on Windows ME.

My first reaction was that the problem results from the following:

If you look at the evaluation for doubling on this move you will see:

Double, pass        : +1.000
Double, take        : +1.117   (+0.117)
No double           : +0.984   (-0.016)

Correct cube action: Double, pass

So the difference between doubling and not doubling is .016. This is
will not be flagged as a doubtful play, so the tutor won't be invoked.

That was my theory. I tried setting the threshold down to .01. It didn't flag
the error.

I just looked at the routine Skill() in play.c.

<Blush>

    case DOUBLE_PASS:
    case REDOUBLE_PASS:

      rDeltaEquity = arDouble [OUTPUT_NODOUBLE] - arDouble [OUTPUT_NODOUBLE];
</Blush>

      rDeltaEquity = arDouble [OUTPUT_NODOUBLE] - arDouble [OUTPUT_DROP];

looks more promising.

How embarrassing.

It would never flag a double/pass error, no matter how grotesque,
since it would always see the equity as 0.

I've just committed a fix for this in the main line code. I think I
should do the 0.13 code as well, as this is really atrocious.

Thanks for your persistance in reporting this error - your example
made it easy to find and fix.

-- 
Jim Segrave           address@hidden