#metabrainz

/

      • mayhem
        chinmay: yay!
      • 2022-11-24 32804, 2022

      • yvanzo
        I don't even remember the last time he’s been on IRC.
      • 2022-11-24 32834, 2022

      • q3lont has quit
      • 2022-11-24 32800, 2022

      • mayhem
        his facebooks suggests that life is more important than being online.
      • 2022-11-24 32809, 2022

      • lucifer
        he recently merged a few on my PRs in mbdata but yeah not much contact other than that.
      • 2022-11-24 32828, 2022

      • lucifer
        !m chinmay
      • 2022-11-24 32828, 2022

      • BrainzBot
        You're doing good work, chinmay!
      • 2022-11-24 32848, 2022

      • mayhem
        when acoustid had troubles, i sent him $500. never head a peep from him.
      • 2022-11-24 32819, 2022

      • BrainzGit
        [listenbrainz-server] release 03untagged-3ae67d7f125034a0fc00 has been published by 14github-actions[bot]: https://github.com/metabrainz/listenbrainz-server…
      • 2022-11-24 32845, 2022

      • lucifer
        uhhh. bad internet connection strikes again...
      • 2022-11-24 32830, 2022

      • mayhem
        need help?
      • 2022-11-24 32842, 2022

      • mayhem could do with a distraction
      • 2022-11-24 32802, 2022

      • lucifer
        thanks but fixed already.
      • 2022-11-24 32848, 2022

      • petitminion
        thank you all :) do you think if I get no response we can implement it ? (We will make shure admins have to opt in an know the service they are using)
      • 2022-11-24 32838, 2022

      • mayhem
        it is open source, so chances are it will be ok.
      • 2022-11-24 32847, 2022

      • mayhem
        petitminion: but I have another thing that you should try.
      • 2022-11-24 32852, 2022

      • mayhem
        its a bit like... magic.
      • 2022-11-24 32810, 2022

      • mayhem
      • 2022-11-24 32828, 2022

      • mayhem
        there is a chance that this could make tagging large collections a lot faster with picard.
      • 2022-11-24 32839, 2022

      • v6lur joined the channel
      • 2022-11-24 32850, 2022

      • petitminion
        mayhem:thank you :) it's to map songs to mbid ?
      • 2022-11-24 32853, 2022

      • lucifer
        yvanzo: great. if bitmap also confirms, i'll start working on removing the wscompat/lucene parts from mb-solr and sir.
      • 2022-11-24 32821, 2022

      • mayhem
        petitminion: yes, by the bucket if the metadata is decent to begin with. should be trivial to try.
      • 2022-11-24 32834, 2022

      • mayhem
        but it doesn't write any thing yet to files. it just shows you a pile of guesses.
      • 2022-11-24 32846, 2022

      • mayhem
        far far far from a complete feature. just a proof of concept
      • 2022-11-24 32805, 2022

      • mayhem
        it uses nothing but the artist and track name.
      • 2022-11-24 32813, 2022

      • yvanzo
        lucifer: 👍
      • 2022-11-24 32821, 2022

      • mayhem
        using release name and track number would make the selection of guesses much easier.
      • 2022-11-24 32838, 2022

      • petitminion
        is it working well ? o/
      • 2022-11-24 32806, 2022

      • mayhem
        I have no untagged music, so I can't really tell on a messy collection.,
      • 2022-11-24 32818, 2022

      • petitminion
        acoustid seem to allow no user interaction and that's what we might want
      • 2022-11-24 32819, 2022

      • mayhem
        but on a collection where the albums are intact, but without mbids, it should shine.
      • 2022-11-24 32841, 2022

      • mayhem
        any solution with no user interaction == spaghetti music collection
      • 2022-11-24 32849, 2022

      • mayhem
        a really really bad idea.
      • 2022-11-24 32811, 2022

      • petitminion
        isn't why acoustid exist ?
      • 2022-11-24 32830, 2022

      • mayhem
        acoustid exists to identify tracks.
      • 2022-11-24 32848, 2022

      • mayhem
        unless I missed it and it now does whole albums. is possible.
      • 2022-11-24 32849, 2022

      • petitminion
        might have errors on acoustif submission but if we combine acoustid and metadata confrontation we should be fine
      • 2022-11-24 32820, 2022

      • mayhem
        petitminion: ok, report back on that when you give up with that plan, please?
      • 2022-11-24 32821, 2022

      • petitminion
        oh no I was thinking of tracks
      • 2022-11-24 32847, 2022

      • petitminion
        why you think it will not work ? ^^
      • 2022-11-24 32810, 2022

      • mayhem
        22 years of experience of it not working.
      • 2022-11-24 32831, 2022

      • petitminion
        okey good to know ^^
      • 2022-11-24 32843, 2022

      • petitminion
        so it should be manual ?
      • 2022-11-24 32818, 2022

      • mayhem
        as automated as possible, but with human review.
      • 2022-11-24 32828, 2022

      • mayhem
        let me give you an example.
      • 2022-11-24 32848, 2022

      • mayhem
        U2's sunday bloody sunday exists on 71 (or was it 91) releases.
      • 2022-11-24 32805, 2022

      • mayhem
        if acoustid tells you its sunday bloody sunday, great!
      • 2022-11-24 32814, 2022

      • mayhem
        which album does it belong to?
      • 2022-11-24 32859, 2022

      • mayhem
        so you have a lot of work to do to work out which of the 91 albums it isn't. then maybe you're down to 5-10 it could be.
      • 2022-11-24 32816, 2022

      • BrainzGit
        [listenbrainz-server] 14amCap1712 opened pull request #2267 (03master…fresh-releases-cron): Add cron job to generate user fresh releases data daily https://github.com/metabrainz/listenbrainz-server…
      • 2022-11-24 32838, 2022

      • lucifer
        mayhem, chinmay, aerozol, monkey, alastairp: https://listenbrainz.org/explore/fresh-releases/
      • 2022-11-24 32802, 2022

      • lucifer
        user fresh releases page is outdated because of missing cron job, should be fixed by tomorrow
      • 2022-11-24 32816, 2022

      • mayhem
        YIIIIISSSSSS
      • 2022-11-24 32819, 2022

      • mayhem does a little dance
      • 2022-11-24 32828, 2022

      • lucifer
        lol spark processed user side in 1 min.
      • 2022-11-24 32852, 2022

      • lucifer
        may become up to date in a few minutes as LB inserts data in db
      • 2022-11-24 32852, 2022

      • mayhem
        plain amazing, lucifer
      • 2022-11-24 32846, 2022

      • mayhem
        !m chinmay & lucifer & monkey
      • 2022-11-24 32846, 2022

      • BrainzBot
        You're doing good work, chinmay & lucifer & monkey!
      • 2022-11-24 32856, 2022

      • mayhem
        I just found out that a band I like has a new EP.
      • 2022-11-24 32800, 2022

      • lucifer
        i had raised the memory limits of cluster last week during experimentation. we are currently running as fast as we can reasonably go. well probably other room for optimization there but have to learn more about various spark tunings
      • 2022-11-24 32810, 2022

      • mayhem
        ding, this feature just earned its keep.
      • 2022-11-24 32810, 2022

      • lucifer
        noice!
      • 2022-11-24 32817, 2022

      • chinmay
        YAAAYYY!
      • 2022-11-24 32800, 2022

      • lucifer
        user page is ingested. reload and it'll be updated as well
      • 2022-11-24 32830, 2022

      • BrainzGit
        [listenbrainz-server] 14amCap1712 merged pull request #2267 (03master…fresh-releases-cron): Add cron job to generate user fresh releases data daily https://github.com/metabrainz/listenbrainz-server…
      • 2022-11-24 32809, 2022

      • lucifer
        mayhem: for artist similarity, i'll create a similarity.artist table on prod with same schema as similarity.recording. sounds fine?
      • 2022-11-24 32826, 2022

      • lucifer
        the spark side of things is mostly done but need to labs api side query to visualize it.
      • 2022-11-24 32848, 2022

      • mayhem
        lucifer: yep!
      • 2022-11-24 32813, 2022

      • petitminion
        mayhem : okey thank you we will think about this :)
      • 2022-11-24 32846, 2022

      • mayhem
        petitminion: if you would like LB (read: me) to help mentor this project, let me know. I bet I could save you a lot of pain.
      • 2022-11-24 32845, 2022

      • mayhem
        auto-tag was basically made for FW. I'm just waiting for feedback from the picard team before I do anything else.
      • 2022-11-24 32835, 2022

      • petitminion
        oooh that so great new !
      • 2022-11-24 32808, 2022

      • mayhem
        :)
      • 2022-11-24 32827, 2022

      • petitminion
        what do you imagine ? Some sort of web UI where fw user could validate tag suggestions ?
      • 2022-11-24 32827, 2022

      • mayhem
        chinmay: lucifer : how would you like to collect feedback about fresh releases?
      • 2022-11-24 32830, 2022

      • outsidecontext
        Regarding this I took a quick look and I think it is very promising. I had a few things about the endpoints I was unsure, but need to look closer at the code
      • 2022-11-24 32854, 2022

      • mayhem
        the endpoints are totally cobbled together with the dataset hoster.
      • 2022-11-24 32813, 2022

      • mayhem
        we'd need to add artist info and then move it to labs.api, but that wouldn't be more than a few hours of work.
      • 2022-11-24 32834, 2022

      • outsidecontext
        We could turn the auto-tag code into a Picard plugin for a quick prototype to play around with
      • 2022-11-24 32841, 2022

      • chinmay
        mayhem: how do we usually do that?
      • 2022-11-24 32845, 2022

      • mayhem
        but this is why I love the dataset hoster. an SQL query and a few minutes of bashing out some guff and you have a perfect exploration page.
      • 2022-11-24 32828, 2022

      • mayhem
        outsidecontext: I think that would be great. but before we do that I think we should brainstorm on how to improve the cluster selection
      • 2022-11-24 32859, 2022

      • mayhem
        ideally, this would be much more AMAZING with release support in the mapping.
      • 2022-11-24 32820, 2022

      • mayhem
        I'm blown away with the possibilities the mapping has brought us.
      • 2022-11-24 32838, 2022

      • outsidecontext
        What I like about auto-tag is that it combines both track and recording lookup cleverly
      • 2022-11-24 32815, 2022

      • mayhem
        I'm still thinking if this would increase or decrease our traffic.
      • 2022-11-24 32821, 2022

      • outsidecontext
        Currently cluster lookup in Picard mostly only looks for releases + matching track count
      • 2022-11-24 32836, 2022

      • mayhem
        wow, it hasn't changed much, I see.
      • 2022-11-24 32800, 2022

      • mayhem
        I think if we get release name support in, then the clusters will be far fewer to look at.
      • 2022-11-24 32815, 2022

      • mayhem
        but this was literally the least data I could start with to see what could work. not shabby.
      • 2022-11-24 32823, 2022

      • outsidecontext
        The clustering itself I'm not sure how much improvement it would bring, that part of Picard works surprising well
      • 2022-11-24 32810, 2022

      • outsidecontext
        But the general lookup approach with these endpoints is definitely the right direction.
      • 2022-11-24 32816, 2022

      • mayhem
        I think the biggest improvement will be in automatically loading more target releases quickly
      • 2022-11-24 32826, 2022

      • mayhem
        I think that will really improve the overall tagging throughput
      • 2022-11-24 32850, 2022

      • mayhem
        general lookup in what sense? how do you see the user using it?
      • 2022-11-24 32819, 2022

      • mayhem
        I envisioned this as an alternate or step of the clustering
      • 2022-11-24 32859, 2022

      • mayhem
        TWO albums from my favorite artists found with fresh releases.
      • 2022-11-24 32809, 2022

      • mayhem
        so much more fun than enduring release radar.
      • 2022-11-24 32817, 2022

      • mayhem
        THREE. 🤯
      • 2022-11-24 32801, 2022

      • lucifer
        mayhem: not sure what you mean? but maybe could discuss here or a ticket i guess?
      • 2022-11-24 32839, 2022

      • mayhem
        I have a lot of UI feedback and feature improvements I would like to see, but I've heard many of them bantered about.
      • 2022-11-24 32851, 2022

      • mayhem
        is there a place that talks about the todo for fresh releases?
      • 2022-11-24 32824, 2022

      • mayhem
        it works pretty well, so no complaints. it needs aerozol love. :)
      • 2022-11-24 32838, 2022

      • lucifer
        i think there was some pending discussion in the merged PR but a ticket would probably be better
      • 2022-11-24 32838, 2022

      • mayhem
        👍
      • 2022-11-24 32854, 2022

      • mayhem
        is it just me or are fresh releases not sorted by date?
      • 2022-11-24 32820, 2022

      • mayhem
        the personalized ones.
      • 2022-11-24 32855, 2022

      • petitminion
        mayhem: if you are interested I opened discussion in the funkwhale forum about matching mbids to tracks https://forum.funkwhale.audio/d/244-acoustid-impl…
      • 2022-11-24 32822, 2022

      • mayhem will look in a sec
      • 2022-11-24 32826, 2022

      • petitminion
        didn't speak of auto-tag since I think you might explaine the purpose better o/
      • 2022-11-24 32842, 2022

      • lucifer
        mayhem: sorted by confidence score i think
      • 2022-11-24 32840, 2022

      • mayhem
        should really be date, otherwise it is confusing with the "all releases" view.
      • 2022-11-24 32854, 2022

      • mayhem
      • 2022-11-24 32854, 2022

      • BrainzBot
        LB-1172: Fresh releases improvements
      • 2022-11-24 32814, 2022

      • v6lur has quit
      • 2022-11-24 32810, 2022

      • v6lur joined the channel
      • 2022-11-24 32857, 2022

      • chinmay
        mayhem: do you want me to change user releases to sort according to date
      • 2022-11-24 32859, 2022

      • chinmay
        ?*
      • 2022-11-24 32829, 2022

      • mayhem
        is the confidence score shown anywhere?
      • 2022-11-24 32850, 2022

      • chinmay
        No :( I can make some arrangements in the card for that
      • 2022-11-24 32858, 2022

      • mayhem
        but, I think it should be sorted by date, since we have a timeline display.
      • 2022-11-24 32815, 2022

      • chinmay
        Yes that makes sense
      • 2022-11-24 32822, 2022

      • mayhem
        what if we displayed the score with the albums, but kept a chronological order?
      • 2022-11-24 32850, 2022

      • mayhem
        something visually simple. 0, 1, 2, or 3 dots.
      • 2022-11-24 32830, 2022

      • chinmay
        A filter I was thinking of was to have the ability to select a date range.. for example users can see releases between 2002-11-20 and 2002-12-19
      • 2022-11-24 32827, 2022

      • chinmay
        mayhem: hmm.. how will users figure dots out? we do need it to be visually simple
      • 2022-11-24 32812, 2022

      • MRiddickW has quit
      • 2022-11-24 32811, 2022

      • mayhem
        petitminion: posted.
      • 2022-11-24 32834, 2022

      • mayhem
        chinmay: what is the confidence score range? is it fixed or open ended?
      • 2022-11-24 32808, 2022

      • chinmay
        mayhem: confidence score is the number of times the user has listened to an artist
      • 2022-11-24 32845, 2022

      • chinmay
        So it can be very skewed range
      • 2022-11-24 32848, 2022

      • mayhem
        open ended then.
      • 2022-11-24 32852, 2022

      • mayhem
        always hard.
      • 2022-11-24 32809, 2022

      • chinmay
        We can normalise it to a scale of 0-10 or 0-100
      • 2022-11-24 32819, 2022

      • mayhem
        but how>
      • 2022-11-24 32823, 2022

      • mayhem
        ?
      • 2022-11-24 32852, 2022

      • mayhem
        we're leaning that that doesn't always work very well.
      • 2022-11-24 32854, 2022

      • chinmay
      • 2022-11-24 32801, 2022

      • chinmay
        Using this formula
      • 2022-11-24 32803, 2022

      • mayhem
        since the scales tend to be non-linear
      • 2022-11-24 32811, 2022

      • chinmay
        oh
      • 2022-11-24 32839, 2022

      • mayhem
        we have some users who listen to the same track on repeat. and others who never listen to a track more than once or so.
      • 2022-11-24 32855, 2022

      • mayhem
        plotting both users on the same scale gives shit results.
      • 2022-11-24 32811, 2022

      • mayhem
        instead we should have a segmented scale like this:
      • 2022-11-24 32820, 2022

      • mayhem
        <5 listens, no dot.
      • 2022-11-24 32825, 2022

      • mayhem
        < 10 listens, 1 dot.
      • 2022-11-24 32834, 2022

      • mayhem
        < 20 listen, 2 dots
      • 2022-11-24 32843, 2022

      • mayhem
        >= 20 listens, 3 dots