I’m wondering what the best/fastest way to train Valence is, so that the tracks picked by Roon Radio which come on automatically after a selection has finished are more appropriate.
I seem to get a lot of choices which, while sort-of vaguely the right thing, aren’t that great. This seems to happen especially with extreme metal music, where lots of the tracks which come up are very obscure, not very good, not quite the right style, or come from a narrow pool of about 10 albums.
I figure things should improve with use, if Valence works the way it’s described. But I was wondering if there’s anything I can do to help it. Make playlists, maybe? Stop automatic suggestions and pick something more appropriate instead? (Or would this just start a new instance of valence choices and not help at all?).
The thumbs up and down are fine, but a bit limited - it doesn’t allow one to make positive alternative choices of material that isn’t coming up via Roon Radio. Plus for example thumbs up seems to be more like ‘play this band/album more’ rather than ‘play stuff that sounds like this rather than the other things you’re bringing up’. And thumbs down seems likely to equivocate, potentially meaning ‘I don’t like stuff like this’ when that could be the reason, or it could actually just be that the track doesn’t quite suit.
I’m not moaning, I don’t expect an algorithm to know immediately about niche music. Just curious how it works, and how I can best use it to make it better.
I’d be curious about this also. I’ve often been pleased with the selections but other times it’s meh, so there’s room for improvement. Last night I found myself wondering if the act of hearting a song that was played, or adding it to a playlist would effect Roon’s subsequent track choices
Yes, I think we’ll all do that. But it’s a bit vague.
Do you mean play music without letting Roon Radio kick in, so it learns in the background? Add up large queues of appropriate choices? Rely more on Roon Radio and then be sure to up/down vote songs one at a time? Add albums from Tidal/wherever to my library, even if I’d be content just to remember them if it weren’t for valence?
Just some general guidelines about how Roon learns best would help.
What I mean is we can’t train Valence. When using Radio we can provide feedback about a particular track for a given seed, but Valance is learning from over 100,000 Roon users who have played millions of hours of music.
So, play the music you love and when radio throws up an artist or album that’s new to you give them airtime too if you like what you hear. Valence will analyse what we play to reveal patterns, trends, and associations relating to our music preferences. I suspect a thumbs up or thumbs down is used only to tweak the algorithms; the important stuff is in the big data, i.e. what we play.
This thread, particularly the explanations of Roon Radio by Brian is really helpful to read, and re-read. It’s full of information and philosophy.
Overall, what you own (your library), what you play (your play history), and what you especially like (favorites) seem to be what we as users control. This almost certainly matters more in some features (Recommend for You, New Releases for You) than others like Roon Radio.
edit: and I should say that ‘Valence’ in this context is more than just Roon Radio.