#metabrainz

ah

mayhem: https://wiki.musicbrainz.org/Development/Priori... done

very good, thanks!

SoC now

datasets are becoming the bane of my existence

but this could be so amazing if it all comes together

what is challenging about the datasets?

just so many cool ideas

I know that feeling, but not in that conext.

context.

mostly that it's very easy to say what we want to do

but actually doing it takes a bunch more

artist filtering is a good example

"artist filtering" - 2 words!

yet, ugly bag of nastiness.

it touches dataset editor, training, access to the mb database

what are you filtering artists on? what is the goal?

the general accepted practise in machine learning is that you shouldn't have more than one example in a dataset by the same artist

because there is the chance of the algorithm learning the style of an artist, or worse, the production style of an album

so the idea is to let someone make a dataset, but when they go to the summary view, we can say "you have 150 items in this class, but once we compare to all the other classes you have we have to remove 48 of them because there are items in other class that share the same artist"

and even there are tricky things. are you going to consider performed as artists? i.e. snoop dogg and calvin broadus in the same set.

you should write a paper about this!

I would put money on the fact that no one has ever considered this

thanks, but... no.

alastairp_: But we consider it all the time?

not sure if this is misplaced optimism, but I think that the data sets that MeB/UPF produce are going to be far more interesting and thought out than other things that exist out there.

anyway, in principal easy, but the live feedback in the dataset editor is somthing I'd really like, which opens up a few more questions

do we use metadata in the lowlevel files? get it from musicbrainz?

merged artist ids are the same thing

Freso: yeah, yeah

;)

yeah, that's point 1.4

we already know that the hacky datasets that we made are better

but they're still not live

zas: wiki http doesn't redirect to https. is that expected?

nope (at least since we decided to move to https), can you create a ticket for it ?

will do

Freso: are we planning on splitting out the ideas page for SoC like last year's one?

if yes, I'd be happy to help you work towards that tomorrow morning

alastairp: I don't know.

alastairp: I didn't do the splitting last year, and I'm not really involved with GSoC this year.

ah! ok

alastairp: yes, that'd be nice.

we just need to lift the stuff from last year's page.

mayhem: done

splitting now

I just copied everything directly. We [that is, not me] will need to edit stuff that's change

I hate mediawiki

someone else can fix picard :)

tsk

alastairp: There you go! SothoTalKer volunteered!

thanks, SothoTalKer!

alastairp, you can always dump mediawiki drudgery on me

on my plate? Idioms are hard.

oh, let's see whoever is faster. i need to prepare some food :)

Hello developers

CallerNo6: you did the awesome ideas page for last year, yeah?

I'd love the same treatment for this year, please.

In fact, it might be good to make a template for us to copy, so we don't reinvent the wheel each year.

copy]

Hi ibrahimsharaf

I've been researching for GSoC 2017 ideas, and I've been interested in AcousticBrainz (New machine learning infrastructure)

I am good with C++, python, and I have basic ML knowledge

I've been playing with scikit learn for some time, solved some kaggle problems

mayhem: I've already copied it

So how can I start?

CallerNo6: this year picard is on the list. I tried to make a nice table with a same-size logo as all the other projects, but I failed

ibrahimsharaf: wow, I just wrote that 15 minutes ago!

we need a logo for the officebrainz project. :)

zome zombiez?

ibrahimsharaf: https://wiki.musicbrainz.org/Development/Summer...

gcilou: ^*3

ibrahimsharaf: we're still working out exactly what we want this project to involve

Freso: I see that as a weird kissyface emoticon o_O

Freso: ?

it's also worth noting that SoC doesn't start for a long time! We have some people interested in working on these projects before SoC starts

Oh

Yeah

Soc?

we'd love for people to participate in our projects outside of the program too

Summer of code

CatQuest: it's that time of year again

arg, stay with oe acronym!

one*

it's always been SoC!

(it's not GCI)

Or GSoc

before people used GosC

oh

erh GSoC

sorry, to me they're identical

ಠ_ಠ

@Freso I'll check the link out, thanks

Google Service on Chip

wow just learned if I shift alt on the - key i get —

oohh noo not the hypens!

That's not a hyphen.

endash

ibrahimsharaf: if you are interested in AcousticBrainz development in general, it would be a good idea to follow the acousticbrainz-specific getting-started guide at https://wiki.musicbrainz.org/Development/Summer... and set up the server too

I've officially given up a long time ago so you might as well not bother :D

the picard logo is bigger, that's why it also is bigger in the wiki

yep, I got that far :)

gcilou was the one with the image editing skills, no? :D

Si

mayhem: do you think we should copy over any of the MB ideas from 2016? I don't remember anyone expressing interest in any of them last year...

some of those even involve more JS than perl

keep "Integrate more *Brainz in more *Brainz" if we can do it without perl.

but it should be enough to scale the picture.

what if we dropped perl from the description and have you do the perl bits? not sure if that is a good balance....

I'd be happy to help out but still dunno what anyone's talking about

hmm...

I've been asleep lol

bitmap: what might be better is to identify things that are pure JS and add those.

right, I'm trying to think of pure JS stuff

since most "features" touch the perl in some way I'm mostly thinking of refactoring-type stuff though :/

gcilou sorry if i woke you up :x

nah it's okay I needed to get up anyway

bitmap: even that will be a hard sel, methinks.

it's only 4 in the afternoon

yeah