#metabrainz

(jekyll can do this automatically too)

if we can come up with some ideas for different categories then it could be OK

I guess, mappings, datasets, demos

Is there a known issue with CB an line-wrapping? e.g. https://critiquebrainz.org/review/81f0aa06-8b9c...

Freso: http://labs.acousticbrainz.org/mappings/million...

I think that last mapping is superfluous. :p

I don't

because the url is the name of the thing

I mean, the last part

it’s not a “million song dataset"

it’s a “million song dataset mapping”

It's also under mappings, so it's implied that it's a mapping already.

yeah, quite possibly

I removed the categories. I quite like just the name of the thing

Aw.

Hi! I'm participating in Google Code In. Is there a separate channel for the competition?

Girish_: Nope. This is it. Welcome! :)

Are the tasks added recently? I didn't see them earlier.

Tasks forMetabrainz.

for Metabrainz *

Freso: hackable how?

CallerNo6: Apparently the author used <pre> or something similar?

alastairp: If I am at http://labs.acousticbrainz.org/mappings/million... (or get the URL handed to me), I may want to cut out the last bit and go to http://labs.acousticbrainz.org/mappings/ to see other mappings.

yeah, that’s a fair point

Girish_: We have plenty of tasks.

chirlu`, yeah, I'm pretty sure it's copy/pasted from blogspot, so funky markup is likely.

Freso: we can probably reverse engineer it in if we need to

Girish_: https://codein.withgoogle.com/dashboard/tasks/?...

alastairp: Likely. :) It's also a young (sub)site, so IMHO it's fine to fumble around for a bit.

Girish_: 47 open tasks right now.

OK, tweeted

hah

from post to like by jherskowitz in 30 seconds

so many notifications

I never get this many likes on my personal tweets

oh, that’s right. because I never tweet

maybe if you added a picture to your account it woudn't look like you're a spammer. :)

doesn't even have to be a picture of you.

I happen to like my egg thank you very much

Freso: and no, I don't think Last.fm have fixed anything, but I found a working API endpoint that can get the user's play counts, and another that can get their loved tracks

nah, they quite seriously broke the output of many apis

oh, huh?

invalid xml, different results

well, the only one I care about was the listening history, which at least still appears to be complete

ok, off to the epiphany parade.

kepstin: the user library API is still completely dead

library.getTracks has been deleted

also one of getAlbums or getArtists, can't remember which

ruaok: have fun!

well, the albums/artists ones weren't really that useful due to lack of disambiguation, and they can be re-created from the listening history anyways

ah, the api you want to use is 'user.getrecenttracks', which is a paginated list of complete scrobbling history including track mbids if submitted

still works afaik, dunno if the format got changed

kepstin: yeah that's what I moved beets over to yesterday ;) I spent some time looking through the API docs, to see what could replace Library.getTracks, and found that one

tjat

that's always the one I've used, since the library stuff was annoying due to last.fm library not being very useful

although annoyingly pylast doesn't give back MBIDs, so I had to derive a custom class that did

and the getrecenttracks of course includes the time of each play, which i don't think the library api included?

in any case, for a listenbrainz import, the getrecenttracks api is far more complete than the current page-scraping method; i guess the only reason it wasn't used was the inability to get a new api key?

(which has since been fixed)

IIRC they were saying they would move to a slower method soon

So I imagine that's what they meant?

But I dunno :)

pretty much the main reason I haven't imported my stuff to listenbrainz yet is the fact that it didn't pull in the mbids from my scrobbles.

(do note that while the 'getrecenttracks' api returns artist, album, and track mbids, only the *track* mbid comes from the scrobble, the others are matched in the last.fm server and are often missing or wrong)

(and the "track" mbid is of course from pre-ngs, so it corresponds to an ngs recording)

mmm, messy data is messy

to be specific, the 'track' mbid in last.fm comes from the 'MUSICBRAINZ_TRACKID' tag (or equivalent in other formats), which in post-ngs picard versions is set to the recording id.

interesting. so acousticbrainz currently covers about 1/4 of the tracks in the msd?

I'm assuming that's probably not intentional, but is just overlap between data people submitted naturally and data in the msd.

Someone is asking details about the solr thing in the forum (in case Mineo or weeksio feel like answering) http://forums.musicbrainz.org/viewtopic.php?pid...

reosarevok, I am not sure that the existing setup is ready for the modifications he wishes to make.

That's a perfectly valid answer for them too I guess :)

kepstin: yeah

honestly, I had hoped we would have more

the matching algorithm is pretty strict though, I think we can easily find a bunch more

though, we use the search server to get the first set of results, so I guess that’s kind of fuzzy

we claim to have almost 2 million unique tracks. In the coming week I’m going to dig into this and try and deduplicate on an artist id/track name level, and see if that’s actually true

recording ids in the same release group, perhaps

because I think part of the reason we have such a low match is that we don’t actually have as many uniques as we thought we did

how are you measuring uniques right now? just unique recording mbids?

might be over-estimating a bit due to unmerged recordings, i guess.

yeah, just recording mbids

which results in weird stuff like https://twitter.com/AcousticBrainz/status/66004...

and of course there's a fair number of similar recordings that'll never be merged, too.

yeah, so track name grouped by artist should give an interesting distribution

recording or track artist ids? :)

uhh. good question :)

I’ll publish a ipython notebook and you can correct it for me!

assuming musicbrainz is perfect, deduplicating by recording id makes sense - as long as you normalize recording ids to handle merged recordings properly.

big assumption tho :)

actually, we don’t handle merged stuff at all either

I wonder how many of the ids we have (from peoples tags) have been subsequently merged

well, fixing that would certainly help your duplicate count

stuff like those beatles tracks have had a lot of merges over time from compilations, etc, and I suspect many users might still have old recording mbids in their files.

Gentlecat, It's working now!! I resubmitted

alastairp: we should split up http://tickets.musicbrainz.org/browse/AB-94 into something more specific

like adding pagination, live editing

yes

and "Evaluating a dataset times out because validation checks that each mbid exists in the lowlevel table." doesn't seem to be related to the editor directly

no, I just made a ticket with a whole bunch of stuff that came up when I was using it

right

how is new schema looking?

I was doing ^ today

merging any time soon?

so we still need to do the same things

dumps, imports, stats, verification of highlevel, conversion

they’re all at the top of my list at work for the next 2 weeks

anything related to datasets directly?

I want to implement something that would allow adding new models into hl evaluation

some kind of admin interface

right

I was looking at getting the merge done as soon as possible, because that dataset stuff depends on having the new schema

so I wanted to merge, and then look at stuff like that

the other big dataset thing is to use lowlevel.id instead of lowlevel.mbid, and allow classless collections (e.g. just like a musicbrainz collection)

stanislas: you around?

reosarevok: yep :)

"Isaac Bashevis Singer (czyta Jerzy Stuhr)" clearly should be two artists - is there a nice way of saying "Isaac Bashevis Singer read by Jerzy Stuhr" in Polish without needing the parens?

semi-colon in between?

drsaunde: no, that would not be correct unfortunately

why not?

reosarevok: I am thinking of something, that might suits you. You could use 'Isaac Bashevis Singer, czyta Jerzy Stuhr'.

it's not hard to do "[Isaac Bashevis Singer] (czyta [Jerzy Stuhr])" as artist credits, we do that all the time for japanese character vocalist artists

I guess, that's better than the parens :)

kepstin: I know, it's just so ugly :p

it's standard formatting in japan, so we just deal with it

reosarevok: I've looked on various polish sites and it is Isaac Bashevis Singer (czyta Jerzy Stuhr) :)

Hmm. Fiiine, I guess I can leave the parens :p

Thanks :)

reosarevok: for example : http://www.gandalf.com.pl/u/opowiadania-dla-dzi...

Did you also happen to find the tracklist for the missing CDs?

Oh, http://www.bemowo.e-bp.pl/bemowo/ini.php?dol=0&... seems to have it

Guess I can fix the stuff then

reosarevok: That library is like 10km from my home.

haha - was the first result when looking for the artists and the track titles