On Tue, Mar 11, 2008 at 8:28 PM, Brianna Laugher
Yes, WHEN we get the infrastructure.
When we see what it looks like and how responsive it is.
When we see how usable it is.
Finding things using category intersections is totally useless when
things have been reduced to a single category, ... and not very useful
when the categories applied are "Bridges made of stone in the Ukraine
with a rated capacity greater than 200 poot"
The idea of 'a category implies its children' isn't possible in high
performance solutions because it would make it possible for a single
user action to have to touch millions of elements of data. (Cat:C
contains all plaints, C is put into A, now the internal references for
A have to have all plants added)... while a pure tagging system is
either O(1) or O(number of cats on an image) operation per update
depending on if the data is forward or inverse indexed.
There is no huge technical challenge in making ultrafast
intersections. Any full-text indexing system can do it. I put up an
example external system built on postgresql some months ago that could
intersect/exclude *any* mixture of categories in a few milliseconds I
turned it off because no one was using it and one day MSN's web spider
decided to enumerate all possible intersections. ;)
I don't see why anyone is going to have much interest in building fast
intersection systems which will have limited usefulness given how many
users believe we should be using categories. ... so if people will
only *consider* possibly changing their behavior after the system is
fully formed and production ready, we seem to have a stalemate.
(Ha, Enwp can keep their inclusionists and deletionists ... At commons
we have intersectionalists and classificationists!)