[mythtv] Paid for Plugin dev?

[f-myth-users at media.mit.edu]

Hmm.  The group I was in at the Media Lab in the early 90's was the
one that invented collaborative filtering, which is the technology
you're describing here.  (It then achieved wider popularity than our
homegrown demos when used as part of a large music-recommendation
system and as part of a spinoff startup, and then places like Amazon
picked up the ball and started to really run with it.)

Even with the technology of 15 years ago, collaborative filtering in
the tens of thousands really wasn't all -that- much computation---at
least, per user.  I'm somewhat surprised that you seem to believe that
it would necessarily be a major resource hog, but on the other hand,
you haven't supplied many details about what sort of recommendations
you think you'd be computing.

This might bear some further discussion at some point, though I'm
not sure that the dev list is the right place.  But if the devs don't
mind, then maybe.

[My assumptions are that you're doing the sort of "people who've
watched -these- shows also watched -this- show, which you have never
watched but is available in your lineup."  Naive scaling of those
algorithms makes them n^2, which is not great but not entirely
hopeless below very large populations and given enough memory to hold
recommendation vectors in RAM---that seems entirely achievable these
days, what with the availability of machines with 4G and up, and
because we don't necessarily have to hold every one of the thousands
of possible shows in any given lineup, but only those that were
-watched- by someone in our cohort cluster.  Back in the early 90's,
it was more like 64-128M, and we still managed hundreds to low
thousands pretty quickly...  Granted, there are only 86,400 seconds in
a day, so if we're talking about that many users, the computation had
better be subsecond if there's only one CPU to throw at it.  SD knows
how many users we have, but has been reluctant (by contract?) to
publicly disclose this value, which is unfortunate here.]