[mythtv] Plans for the new 'recordedprogram' and related tables?
brad+mydev at templetons.com
Wed May 18 08:42:39 UTC 2005
On Wed, May 18, 2005 at 09:12:45AM +0100, Mark wrote:
> Brad Templeton wrote:
> > I'm doing a Tivo-like "suggestions" system which wants to train itself
> > by looking at the genres, credits and titles of old recordings, along
> > with explicit user commands.
> A couple of people have taken stabs at this before, check the lists.
> One was Steve at nexus uk who wrote a myth-bayes module, forget who did
> the other one. It may save you reinventing the wheel.
> Haven't tried either myself
> ahh, found it, it's quite old tho
For myth 0.12, it's going to be pretty out of date.
However, mine didn't take long to write, it's already roughly working as added
commands in tvwish. Once I built the basic libraries in tvwish to deal with
mythtv's database in perl, the result is the ability to script mythtv recording
scheduling in perl, where one can code these sorts of things way faster than C++
I'm not super thrilled with the bayesian algorithms on keywords from descriptions
approach, or the things Tivo does which I think includes a bit of that but more
genres and actors. Even as inaccurate as it is, the Tivo approach is way more
useful than you first think it would be at first. It becomes a replacement for
surfing random shows, except your set to surf is far less random than "what's on
now" but far more random than your season passes.
And now myth's tools are sophisticated enough to deal with the suggestions existing
with super low priority and super high auto-expire, so that they don't interfere
with any real recordings. To be truly invisible, like tivo's, you would want the
live-tv system to happily kill a suggestion (anything below -100 priority, say) if
a tuner is desired for live tv, and likewise not show these very low priority recordings
(suggestion or otherwise) in the upcoming list unless asked etc.
More information about the mythtv-dev