Yes, this is the sort of thing I had in mind. We seem to be thinking along
the same lines in terms of using Metakit to support knowledge exploration.
I have this concept of a "knowledge bank" that is closed under "knowledge
operations", and a "knowledge pathway" that the user can navigate and
modify.
In terms of efficiency, I'm not sure whether it is better to actually keep
the views or keep the operations (like an edit history). This depends
partly, of course, on the expense of each; and perhaps the best approach is
to support both.
But I do believe that Metakit is a powerful tool to use in such a context.
--------------------------------------
Gary H. Merrill
Director and Principal Scientist, New Applications
Data Exploration Sciences
GlaxoSmithKline Inc.
(919) 483-8456
"Brian Kelley"
<[EMAIL PROTECTED]
du>
To: [EMAIL PROTECTED]
26-Feb-2004 11:07 cc: [EMAIL PROTECTED]
Subject: Re: [Metakit] Saving views for
recreation later
[EMAIL PROTECTED] wrote:
>I strongly second Brian's request/suggestion!! -- I think. This would
be
>an extraordinarily useful feature for using Metakit in a datamining
>environment where you want to keep a "trail" of results and refinements --
>which is exactly what I'm involved in implementing.
>
>But if what you end up with is just a view, can't you simply save that
now?
>I haven't yet got to the stage where I've actually tried to save the
stages
>of an exploration like this.
>
>Or is Brian asking for some additional bookeeping to keep track of such
>views and make the process simpler?
>
I'm doing the bookkeeping myself now, each view contains the history of
the operations that were performed on it. This includes joins and the
like.
vw.history() -> yields something like
['vw1 = st.view("table1")',
'vw2 = st.view("table2")',
'self = vw1.join(vw2, vw2.id)']
Which I can save so that the table I'm looking at could be regenerated.
This turns out to be easy to implement using my wrapper classes.
Actually, one thing that is missing in the python wrapper now that I
think about it is the function that allows a view to be copied into a
view with the same structure. I think this exists in C++. I wonder if
it would be useful to have a clone function for a view such that:
clone = vw.clone(name="temporary_table")
where a new table "temporary_table" would be generated in the database
and filled with visble rows/columns in vw. This would be a way of
saving some form of state.
Brian
Brian Kelley
_____________________________________________
Metakit mailing list - [EMAIL PROTECTED]
http://www.equi4.com/mailman/listinfo/metakit
