Intelligent Recording selection?

[pjpjpjpj]

Late chiming in, and just had a thought. In all of the sifting through the open source code, has anyone looked at how Sage comes up with intelligent recordings? That algorithm would be interesting to look at, especially in light of trying to determine a show's "worth" to you... in other words, it would be interesting to see how it judges the "worth" of a show as far as using it to come up with similar shows to intelligently record.

It wouldn't be a true determiner of what you watch, but it would be interesting to see.

* split from another topic *

[Skirge01]

Originally Posted by pjpjpjpj

Late chiming in, and just had a thought. In all of the sifting through the open source code, has anyone looked at how Sage comes up with intelligent recordings? That algorithm would be interesting to look at, especially in light of trying to determine a show's "worth" to you... in other words, it would be interesting to see how it judges the "worth" of a show as far as using it to come up with similar shows to intelligently record.

It wouldn't be a true determiner of what you watch, but it would be interesting to see.

Getting off topic here, but I'm hoping some awesome dev will decide to take it a step or two further and integrate similar interests across users, perhaps utilizing something like Trakt. I have that plugin for Kodi and it's incredible (and a little scary) how much data they gather.

[Opus4]

One of the recent core + STV changes I submitted includes the ability to show on the Detailed Info screen what IR agent caused an IR show to be scheduled/recorded, if the option is enabled in Detailed Setup -> General.

What I really wanted it for was to look into why Hopalong Casidy kept getting recorded even though I kept telling it not to. Part of the tweak was to cause Don't Likes to have more weight than they used to & thus to solve this particular problem.

Andy

[Fuzzy]

Originally Posted by Skirge01

Getting off topic here, but I'm hoping some awesome dev will decide to take it a step or two further and integrate similar interests across users, perhaps utilizing something like Trakt. I have that plugin for Kodi and it's incredible (and a little scary) how much data they gather.

For the most part, it's a simple scoring system, that takes in user created flags (Don't Like), watched history, and similarity. Shows that IR records and are not ever watched decrease similar results - so you don't HAVE to 'don't like' to train the system, but don't like flags are stronger than 'wasted' flags (as far as I can tell).

[Fuzzy]

Originally Posted by Fuzzy

For the most part, it's a simple scoring system, that takes in user created flags (Don't Like), watched history, and similarity. Shows that IR records and are not ever watched decrease similar results - so you don't HAVE to 'don't like' to train the system, but don't like flags are stronger than 'wasted' flags (as far as I can tell).

Now, of course, this is all quite 'simplified', as in the background, it actually uses these types of scoring to track similarities and create 'Agents' which are mechanically what favorites, that will do things like record a certain category of show on a given channel, shows of the same category with the same actor, etc. The v9 STV that Andy mentioned does have the ability to view the agent that recorded a particular show, but since I use gemstone, I'm not using the v9 SageTV7.xml

[troycheek]

Back when I still used my ISP's email, I loved a program called POPFile which classified incoming messages using a "naïve Bayes algorithm." While it was designed to be a general sorting system for incoming mail (work, personal, mailing lists, etc) it also worked as a simple spam/not spam filter. With very little training, it got scary good very fast. I don't pretend to understand how it works, but it works well. Often, when SageTV's Intelligent Recording would record something completely unlike anything I'd ever watched (but very similar to shows I'd repeatedly marked Don't Like), I wondered how well POPFile's algorithm would work with SageTV metadata.

Hmm. Given that I've recently gotten the Web UI working again, I wonder if I can export a bunch of metadata for shows I like and don't like and mail them to myself...

[Narflex]

I can concur that Fuzzy is giving an accurate explanation.