John Walsh writes:
                We can't just have a single tune-search algorithm.  That's not the
         way this group operates. :-) ...

Heh. So true.  But it's also based on differences of what people want
to  do.  I've been reading the messages on the subject with interest,
despite the fact that they are far from suitable for the things  that
I'd like to do.

The  reason is simple:  I've been looking at it from the viewpoint of
my "tune finder", which would take a (fragment of a) tune from a user
and  try to find similar tunes among the 50000+ ABC tunes on the web.
This imposes some constraints that  the  described  algorithms  don't
seem to deal with.

The  main  problem is having the user wait around for the comparisons
to complete.  My tune finder is based on a background search bot that
takes  about  two  days  to run.  This is due to the delays in making
Internet connections. The searcher uses very little cpu time, because
it  spends most of its time waiting for a TCP connection to complete.
Adding a tune comparison to it would be straightforward. But if users
have to wait two days for a reply, it simply wouldn't get used.

The alternative seems to be downloading all the ABC on the web into a
database. This isn't feasible in my case; I would definitely wear out
my welcome here if I were to do such a thing.  As it is, there's only
one  other  user  that  uses  more  disk space.  The machine's owners
approve of my use of their machine, as long as it's  not  commercial,
as you might expect of an MIT department.  But there are limits.

Anyway, even if I thought I could get away with downloading  all  the
ABC  in  the world, I wouldn't do it.  An interesting thing about the
web is the way it  encourages  distributing  data  across  a  lot  of
machines. I'm mostly interested in encouraging people to set up their
own little collections of tunes, and seeing what can be done with the
resulting collection of collections.

So far, the discussion of how to match and compare tunes doesn't seem
to be at all practical if the tunes are scattered across the web. The
algorithms would take days, and thus would be useless to users.

(But don't stop discussing the subject.  We might be able to do  some
interesting things with a subset of the available music.  ;-)

To subscribe/unsubscribe, point your browser to: http://www.tullochgorm.com/lists.html

Reply via email to