#metabrainz

Pratha-Fish: https://mastodon.social/@metabrainz/11140031712...

kellnerd: https://mastodon.social/@metabrainz/11140025398...

And on X and Bluesky as well. Sorry it took a bit! I wanted to read the posts properly first :)

Yayy :D

About to go to bed, but before I forget: ansh, I’ve heard good things about the Fresh Releases update from monkey, but can I see it before go live? I don’t mind monkey signing off on anything and everything, but because I did the mockup for this one I’d like to check that nothing has been lost in between :)

bitmap, reosarevok: I did draft a blog post for today's release and have a local branch for MB Docker Compose, not pushed yet ^

mayhem: I will try be on at your 7pm for our catch-up by the way, hope that works for you!

Night all xo

thats meeting time. an hour later.

nn

Oh wait, has daylight savings come back round again? Wow, I could start attending meetings again!

ding. :)

lets do that and then you and I have a chat after.

Mailed in my review anyway, but might be there :) see you at 8

great!

Have a great Monday all. A challenge, but you can do it

Moin!

mayhem, lucifer: turns out I'm not the only user that has been pigeonholed with a single band in their recommendations, here's someone else who also has 100% King Gizzard: https://beta.listenbrainz.org/user/silverbluep/...

Still not sure what's going on, but there's an overfitting issue somewhere

moin!

if that other user was pigeonholed by another artist, I would be worried.

but king gizzard is so prolific it messes with the alg.

Hah, that's wild.

Is it because there isn't enough "similar" artists based on KGLW?

Because it's not like either of use only listen to this one band.

us*

hah. even *I* have one song by king gizz

hm, is it a script or have we added "submit votes and edit notes" button on the top of pages too?

(becasue that's geat and something I'vewanted forever)

In both cases KGLW is significantly more listened to compared to other bands, but we do have other bands to pull from: https://usercontent.irccloud-cdn.com/file/C5n80...

haha weedian

i dont think it has much to do with popularity, but about the number of songs they have and keep releasing. the alg constantly think your're behind in listening to them.

yvanzo: thanks! I've also been busy until now and today is weird but we release it all in the evening if need be

Hey, does the "/1/cf/recommendation/user/{user/name}/recording" API endpoint only base recommendations off loved/hated songs?

Do regular listens have no effect on the recommendation algorithm?

LumePart: no, those are not taken into consideration yet.

Okay, thanks

we manually remove hated tracks if they system suggests to include one. before too long we will need to improve this and actually have the model take the love/hate into account, but I am uncertain how to do that. so far I've not been able to get good advice on this.

mayhem: I checked and all the tracks that appear in my created for you playlist have been listened to before, while there are some other tracks in my raw recs that I haven't listened to before. If the alg thought I should listen to the KGLW I haven't listened to yet, that would be a positive, but it seems to only select tracks I have listened to before.

🤷‍♂️

also, remember that the recommendation has a weight associated with it. it is all about the weight, not about if you've listened to it before. for some tracks it will think it is more important for you to re-listen to a track than to listen to a new track.

OK, that might help narrow down where the issue is, then. Not sure how it applies those weights, but it's currently broken

in https://listenbrainz.readthedocs.io/en/latest/u... there is a feedback import endpoint, but it is mostly undocumented. Can this be used or is it not meant for external use?

There's also a "latest_listened_at" field in the response, which could help disrecarding the most recently listened tracks (i.e ignore all songs that have been listened within 1 month)

monkey: thats the collaborative filtering algorithm, which is a black box to us. hard to say if it is broken or not. I am not an expert enough to say.

Is there anything I can do to help investigate?

I don't know where to start. literally, to me, this is the painpoint of ML. At some point you get bad results and want them to be different, but... you can't kick the algorithm. you can't readily tune it.

Happy to help any way I can, it's been months since I've had a recommendation playlist worth anything and I suspect now that it's not an isolated case. It would be nice to prioritize fixing this important feature

I see

I wish I could help, I wish I understood more, but... I simply don't.

Perhaps we can focus our efforts as a team to try and tackle this

what would help, i think, is if you found another user and another band that exhibits this problem.

monkey: perhaps. go ahead, lead the charge.

Lol, found somehow else in the KGLW black hole :D https://listenbrainz.org/user/jakestanger/recom...

do you know of other very prolific artists? I would find one and see if that artist exhibits the same problem. that would be a good clue.

Oh dear. I keep finding more users with the same issue, all related to KGLW

when I get back from india, I can see about getting on another call with Normal Casagrande to try and get ideas how to solve this.

*norman

story of my life, lol.

Lol

This user has the same issue with The Beatles: https://test.listenbrainz.org/user/freakanth/re...

ahhhh, ok. that is interesting.

this makes me wonder -- when we fix the beatles problem, will that fix/impact this problem.

Norman Casagrande? The one who was working at Last.fm?

I found some similarly skewed playlists with Pink Floyd, Radiohead and Aphex Twins. I wonder if the issue is the ratio of listens from one artist compared to other listens, or if it is indeed related to the popularity or prolificness of these artists

[listenbrainz-server] 14MonkeyDo opened pull request #2626 (03master…art-creator-copy-button): Fix art creator copy image funcionnality https://github.com/metabrainz/listenbrainz-serv...

Pratha-Fish: that one.

And if I am comprehending this chat correctly, he's also helping out with the music recommendation systems in LB?

Any machine learning, recommendation stuff he has agreed to help us, yes.

That dude has some jaw dropping credentials to his name. It's kinda exciting to have him helping out at MeB 😳

:D

yvanzo: thanks, I'm reviewing the blog post now

bitmap: we should probably get the PR #3080 in today’s release, is there any other needed PR?

I don't think so. I'll have a look at your comments

yvanzo: did you see https://github.com/metabrainz/musicbrainz-serve... ? musicbrainz-docker will have to migrate too

yes

good, just making sure

actually I only managed the `corepack enable` part

but I found out that my `yarn.lock` is having local changes now :/

I think I didn’t pay enough attention to the changes that come with Yarn 4.

I hoped it would be seamless, though we should add some brief instructions to the blog post for people not using Docker.

what sort of local changes are you seeing?

[musicbrainz-docker] 14yvanzo opened pull request #262 (03master…node-20): Upgrade MusicBrainz Server to Node 20 & Perl 5.34 https://github.com/metabrainz/musicbrainz-docke...

bitmap: yes, some brief instructions (such as ditching old Perl packages) would be nice

bitmap: nvm about local changes, it’s just that I did run an old image in the meantime.

I don't recall any Perl packages being dropped intentionally, so perhaps it was a side-effect of regenerating cpanfile.snapshot?

ah, okay

for the Perl packages, it seems to be due from upgrading to Perl 5.34.

s/from/to/

ah, I see mbdocker has a separate cpanm step after the main --installdeps one. I haven't determined if those can be removed on other systems

It’s not a problem for mirrors though, as the whole image is rebuilt, it’s just for development setup.

[musicbrainz-server] 14mwiencek merged pull request #3080 (03master…mbs-13336-improvements): RebuildIndexesUsingCollations.pl improvements https://github.com/metabrainz/musicbrainz-serve...

bitmap: Thanks, updating beta…

Thank you for sharing the GSoC blog posts on our socials aerozol, your summary pretty much nails it :)

yvanzo: thanks. I just added a paragraph about Yarn to the blog post. sadly, I can't remember how to format code blocks in Wordpress

never mind, I found it. (all I had to do was complain)

lucky you

bitmap: does `node_modules` really have to be removed?

reosarevok at least ran into an issue on his machine where removing it resolved things

it was "Usage Error: Couldn't find the node_modules state file - running an install might help (findPackageLocation)" while he was trying to upgrade a package via `yarn upgrade`

perhaps there is a less nuclear option to resolve that

mfw i am trying ot add an ainstrument after 2 weeks and it's right in the beta release :D

Lols

wassat?

bitmap: MB Docker release draft is up for review

aerozol: Definitely! We'll deploy the PR on test.LB so that we can get a feedback from you as well before merging the PR.

ansh: Want me to rebuild and redeploy the FR feature? I've got LB#2626 up at the moment, but it's quick to test the feature in question

Fix art creator copy image functionality: https://github.com/metabrainz/listenbrainz-serv...

Sure

yvanzo: looks good to me

there is nothing wrong with irc

mayhem ApeKattQuest I am collecting feedback from new contributors interested in contributing to MetaBrainz, and almost all of the new ones are either using kiwiirc or element.io to communicate, and are facing a TON of issues due to the same.

I mean, that sounds like a client issue

You're right.

Kiwiirc doesn't sync chatlogs after disconnecting.

element.io *was* the only free and viable client with chat sync, but the libera.chat and matrix IRC bridge got nuked.

the matrix bridge really needs to be made up again