From: Warren Brown <wkb@...>
Date: 2003-02-15T02:14:08+09:00
Subject: Re: Directed graph quiz application

Brian,

> I was thinking about something like this:
> atomic number:name:symbol:family
> 1:hydrogen:H:nm
> 2:helium:He:nm
>
> and so forth, a little like CSV.  But the part I'm
> confused about is that the questioning system should
> work on a directed graph system, as in some questions
> might be
> 1's name?
> 1's symbol?
> 1's family?
> hydrogen's atomic number?
> hydrogen's symbol?
>
> and so forth.  However, I *can't* ask
> nm's name?
>
> because the designation 'nonmetal' shouldn't point back
> to the other fields.  It is a one way mapping.  So I
> think I have to establish a directed graph system
> between the fields so that the quiz application knows
> what questions to ask.

    The trivial answer is to change your format to question:answer and then
populate the database with only valid combinations.  This is easy to code
but requires a lot of typing and redundancy.  A better solution would be to
have your questioning system do this for you.

> I'm confused about how to describe these mappings in the little quiz
> database files.  It'd be nice for it to be human-editable and stuff.

    As far as a database syntax, I would use something like:

=atomic number=name=symbol<family
1:hydrogen:H:nm
2:helium:He:nm

    This format would allow you to easily recognize "description" lines
since they start with '=', and would readily show which attributes are
one-to-one ('=') and which are many-to-one ('<') in relation to the base
(first) attribute.  It sounds like these are the only two relationships that
you are worried about, but you could add others.

    Internally, you would use the "=atomic number=name=symbol<family" line
to create an array of attributes (or columns):

attributes[] = [
  ['atomic number','='],
  ['name',         '='],
  ['symbol',       '='],
  ['family',       '<']
]

    If all you want to accomplish is to *eliminate* the many-to-one
questions (e.g. "nm's name"), your questioning system could convert each
"data" line read from the database in into multiple question:answer pairs.
For example:

1:hydrogen:H:nm => 1:hydrogen
                   1:H
                   1:nm
                   hydrogen:1
                   hydrogen:H
                   hydrogen:nm
                   H:1
                   H:hydrogen
                   H:nm

    Any attribute with a many-to-one flag does not appear as a question.

    Of course, if you wanted to get really clever, you could actually
collect these many-to-one attribute "answers" together and then ask
questions like "What are the names of all nm's?" and "What are the atomic
numbers of all nm's?"  Unfortunately, this gets a little more complicated.

    Hope this helps!

    - Warren Brown