RE: [Bug-gnubg] Match statistics graph

[Albert Silver]

I've been following this and agree about the danger of a bloated
program. And just so you guys can get a good laugh, that program I
worked on, Chess Assistant, gained some useless e-mail client in it (as
spartan as could be) to send and receive bases and games. Very much to
my protestations. However, I don't agree with the restrictions being
imposed here. I agree there should be some, and in that case, let it be
at the user's discretion to create new graphs, but we're not there yet
IMHO. 

I weigh two factors when deciding whether to opt for a new feature:

- Amount of code/work involved compared to usefulness of function. 

This is obviously from a developer's point of view. To put it simply: is
it worth the effort?

- Usefulness of function to user. 

The usefulness isn't only a matter of the user using it, but also
applies to the suggested 'let the user do it'. Would enough users who
would find this useful know how to do it? If not, the argument is
pointless since by choosing this, we know in practice it will simply
never be taken advantage of. This is similar to, but not as bad as,
suggesting that if a user wants something they can 'simply' use the
internal Python to script their desires. My only reaction would be that
you might as well say that apart from the developers, maybe a small
handful of people (at most) will even be capable of doing this.

This last one is key here. How many graphs are useful enough to be
included as a courtesy of usefulness and ease-of-use (sure beats
exporting and building). I think some graphs would be useful enough to
deserve implementing directly here, but that many would not. The
question would be more which ones, which would be a combination of group
decision and of course the person implementing them.

Snowie has a number of graphs, and still not enough IMHO. It has
breakdowns of the equity per match (in its Player Records) and per game,
of the cube decisions, the checker play, as well as the luck. If one
really keeps a good record of results, so that enough information is so
recorded, then it would be possible to get a greater breakdown of
results, and yes, I think this would be very interesting. I also think
enough serious students of the game would find it interesting to make it
worthwhile: The error rate at different scores (only for 5-point match
scores which are the most crucial), and the error rates in games at
those scores. This would not need a graph since the graph wouldn't be of
much help, but the information is interesting. If I saw that my error
rate for DMP games, or 1-point games, was higher than it should, I'll
know I have a weakness there. A beginner might not care or understand,
but a more advanced student most certainly will. Some other critical
scores such as 2-away 4-away, the ultimate Gammon-Seek and Gammon-Avoid
score are also interesting. I have found that a common score where
strong (1800+) opponents consistently make doubling blunders(!) is
4-away 3-away. The break down would be fairly simple once the
information was available, and would probably look something like:


Scores          Checker Error rate   Wr.Dble  Missed etc.   No.Games
5-away 5-away        0.012                   0.07      0.128         4
5-away 4-away           Blablabla  (you get the idea)
Money games                     More Blablabla

'All' that would be necessary would be for it to keep a log of these
details when saving. I think enough users would find this useful for
this to be of interest, and it would provide a much clearer picture of
what we are doing right and wrong. Playing at 2-away 4-away is VERY
different than at different scores, and who knows? Maybe I'm much weaker
at this and don't play aggressive enough for example. Naturally, I
wouldn't impose this as an immediately available possibility, it would
be offered as a choice for a "Detailed breakdown of error rates". 

Just my opinion.

                                                Albert

> On Fri, Aug 29, 2003 at 09:51:41AM +0000, Jonathan Kinsey wrote
> > I agree with the emails about avoiding making gnubg do everything.
> >
> > It would be nice to add some export of data functions for expert
users.
> On
> > the other hand it's good to add generally useful things which
everyone
> will
> > use.  Not sure the graph falls into the catagory though :-)
> 
> Yes, I think it's fine that we supply a single graph.
> 
> But just two weeks from now you'll have 10 users requesting 10
different
> new graphs. We might as well anticipate this and make an export of the
> raw data, so we can say to the user: "you can export the data and make
> your own graph". The user will reply: "but this particular graph is of
> interest for many users -- please implement it", to which we *must*
> reply: "yes, but we only supply this single graph". Otherwise we'll
end
> up with a bloated gnubg that includes an email client for sending
> positions and matches to other users (a gnubg specific corollary of
> Zawinski's law).