[mythtv] Paid for Plugin dev?

[Erich Nachbar]

> Can you give us some of the numbers your've come up with?  *.mythtv.org
> runs on our own hardware.

I totally can. Maybe someone can give me an idea of how many
subscribers we are talking about (e.g. Schedules Direct subs) so we
can be more realistic? For all I know my assumptions can be totally
off.

> Were you thinking of something that uploaded recorded activity
> automatically in the middle of the night and downloaded recommendations
> or would it be user initiated and generate recommendations on demand?

The whole process would work similar to the schedule update. The
upcoming show schedule is known 2 weeks in advance making it ideal to
be downloaded nightly.
I don't believe anyone computes recommendations for a large user base
on demand due to the involved data volumes & runtimes.

> Not speaking for the Schedules Direct guys, but it could be something
> that could make sense for them to support if necessary.  It all depends
> on what kind of hardware footprint we're talking about.

I'll publish some numbers once I can get a better idea of how many
subs the system should be designed for.

> Other questions/ideas.  During the upload process, would you want the full
> recorded history or just things that were recorded since last time?  Would
> users have to register with this site or would it be automatic by anonymous
> ID?  Can you describe a little more about the process you're
> envisioning?

The nightly upload process would ideally send only incrementals of
your watched shows. On the server we will need to keep some history
around to find similar users and make recommendations. It could be
through an anonymous ID if there is a way to pay for the required
infrastructure.

As part of the upload process of the viewing history the server sends
down the computed recommendations. The recommendations have been
computed based on the prev. nights shows (i.e. lag one day behind).
Due to the fact that we already know the upcoming shows in advance
this shouldn't be a problem. Once the recommendations are available on
the backend, one could come up with ways to present that to the user.
I have not spent too much time (yet) thinking about a great way to
interact with recommendations.

Ideally recommendations are based on as much data as possible including:
- largest possible user base to find the best match (like you already mentioned)
- as much user input as possible (user's current scheduled shows,
watched shows, show ratings, deleted schedules, deleted shows, kept
shows, etc.)

The number of computations are rather large and would not be done in a
typical RDBMS (unless you wanted to wait weeks to be done with one run
or spend $$$$), so the required infrastructure is very different from
what people are used to (aka LAMP).