> Why things work the way they do:
>
> a) it is desirable to analyse the computer player as you go along,
> so as not to waste time during analysis after the game.
> b) it is desirable to keep the analysis and the annotation in sync.
> c) a double/no double analysis has to include a take/pass analysis
> to get the right action
> d) the take analysis belong with the double analysis so as not to
waste time
I do agree without hesitation on points b, c and d.
Not really on point a. Users may want to have gnubg playing at, let's say,
World Class since nowadays it's fast, but they may want to have the anlysis at
GrandMaster. A slow analysis after the match is probably affordable,
but not slow play.
> For the computer the following is done
> before roll:
> a 0-ply assessment of our position in the double window. If this is
> false no further double analysis is done, if true a full n-ply
> double/no double // take/drop analysis is done.
> after roll:
> luck determination
>
> For the human player analysis is only done when the tutor is on.
>
> So what get's annotated when tutor is off is
> A1) computer luck
> A2) computer errors if the decision differs from a stored evaluation
> B) player take/pass decisions (annotated, but not interrupted)
> B1) Any decision where analysis is stored. That is if you for
> example do hint on a chequer move and still choose an inferior move,
> it gets annotated, but not interrupted.
>
> What has been changed is b)
>
> What I suggest when tutor is off:
>
> a) not doing luck analysis
> b) not storing the double analysis obtained from the computer move
>
> It will cost a bit of time when analysing, but should otherwise be
ok.
To me, no tutor should mean no analysis at all (neither
luck nor cubes, in or out of the doubling window).
Thinking about all this stuff raised a few quetions/remarks
in my mind:
1. I think I've already reported this a while ago:
in the options window (settings/options/tutor) I have a draw bug:
the frame "Tutor decisions" is too small and/or empty. See it
here:
2. Actually the whole tutor thing is in fact the union
of 2 different things: "on the fly analysis" (decision
by decision instead of at the end of the match), plus the warning mechanism in case
of errors (only applicable for human players, of course).
"on the fly analysis" is, to me, a feature
by itself: it slows a bit the playing pace but at the end of the game/session/match
the analysis is already available, no wait to know your error rate.
I would go for:
1 option for "on the fly analysis"
(both players, whether gnubg or human) Values: none /
same as eval / same as analysis 1 option for tutor (on human players) values: warn on
doubtful / bad / very bad / never
Tutor option disabled if "on the fly analysis"
is none.
Minor remark: when "on the fly analysis"
is ON, as soon as a game (or the entire match) is over we should have all the stats
just like if an entire analysis is run: per game (and per match, if
over) totals, error rates etc. If "on the fly analysis" is ON,
there should be no need to (re)analyze the match/session/game again.
3. A final word about terminolgy: currently it's a
bit misleading. The "Evaluation" settings act on the Hint
command. My proposed "on the fly analysis" option
can be set as "same as Analysis" or "same as Evaluation". Weird.
I would do the following:
- rename the current "Settings/Evaluation"
to "Settings/Hint" - rename my "on the fly analysis" as "on
the fly evaluation", with possible values none / same as hint / same as analysis