but bitmap said that is intended and the inclusion of the same in XML output was an unintended side-effect (however it is used in some of the picard code)
2017-07-15 19647, 2017
samj1912
but the thing is, I can't see any part of the main code using that parsed output
zas: I think I addressed all the comments on that PR
2017-07-15 19642, 2017
samj1912
we will need to test all of this throughly though
2017-07-15 19655, 2017
samj1912
I will be writing more tests soon
2017-07-15 19604, 2017
samj1912
but meanwhile, any manual testing is appreciated
2017-07-15 19642, 2017
SothoTalKer joined the channel
2017-07-15 19629, 2017
Gazooo has quit
2017-07-15 19655, 2017
Nyanko-sensei joined the channel
2017-07-15 19605, 2017
D4RK-PH0ENiX has quit
2017-07-15 19604, 2017
SothoTalKer has quit
2017-07-15 19644, 2017
SothoTalKer joined the channel
2017-07-15 19603, 2017
Nyanko-sensei has quit
2017-07-15 19639, 2017
agentsim has quit
2017-07-15 19650, 2017
D4RK-PH0ENiX joined the channel
2017-07-15 19608, 2017
D4RK-PH0ENiX has quit
2017-07-15 19651, 2017
sagar-kohli[m] has quit
2017-07-15 19607, 2017
D4RK-PH0ENiX joined the channel
2017-07-15 19610, 2017
psolanki has quit
2017-07-15 19627, 2017
SothoTalKer has quit
2017-07-15 19656, 2017
Guest7249 joined the channel
2017-07-15 19656, 2017
sagar-kohli[m] joined the channel
2017-07-15 19604, 2017
Mineo has quit
2017-07-15 19604, 2017
QuoraUK has quit
2017-07-15 19616, 2017
ZarkBit joined the channel
2017-07-15 19640, 2017
SothoTalKer joined the channel
2017-07-15 19621, 2017
MajorLurker has quit
2017-07-15 19649, 2017
SothoTalker_ joined the channel
2017-07-15 19650, 2017
SothoTalKer has quit
2017-07-15 19606, 2017
lazka joined the channel
2017-07-15 19646, 2017
samj1912 has quit
2017-07-15 19627, 2017
magerharz joined the channel
2017-07-15 19646, 2017
cjnolet
Is listenbrainz only using last.fm data that actually has MBIDs or are you guys doing somethign special to match up the noisy data (without mbidas) to MBIDs?
Reason I was asking about the ListenBrainz mapping to MBIDs is because that noise could cause severely detriment the accuracy of recommender models.
2017-07-15 19641, 2017
cjnolet_work
For instance, I just scraped a bunch of last.fm information for several users that I had collected and was looking through the database (I'm using Elasticsearch) and found over 100 different Madonna's in the database that didn't match up whatsoever.
2017-07-15 19601, 2017
cjnolet_work
Because some were listed as "Madonna ---- http://altunes.com" while others were listed as "**** MADONNA ****"
2017-07-15 19658, 2017
cjnolet_work
Problem is, the people who had those in their taste profile won't match up as liking Madonna, they match up as liking that variation. But I'm also trying to use my database as a type-ahead for users in an app and "Madonna ---- http://altunes.com" isn't a pleasant thing to have pop up, lol
2017-07-15 19615, 2017
cjnolet_work
That's the only reason I bring it up.... it's noisy and it's actually hindering.
2017-07-15 19621, 2017
cjnolet_work
vs. just dropping those records.
2017-07-15 19618, 2017
cjnolet_work
Though that's just my opinion. I was curious if any othe rmembers of the community thought differently and perhaps there's a larger reason why that I could use to my benefit
2017-07-15 19610, 2017
UmkaDK has quit
2017-07-15 19618, 2017
UmkaDK joined the channel
2017-07-15 19612, 2017
UmkaDK_ joined the channel
2017-07-15 19630, 2017
UmkaDK has quit
2017-07-15 19639, 2017
alastairp
cjnolet_work: so, we created a thing called messybrainz
2017-07-15 19653, 2017
cjnolet_work
Ah, okay, that's what that is
2017-07-15 19657, 2017
alastairp
the idea is, that every string artist gets given a different uuid (messybrainzid)
2017-07-15 19614, 2017
alastairp
the minimum data needed for a messybrainzid is an artist and a track name
2017-07-15 19633, 2017
alastairp
but if the listen/scrobble contains album, track position, etc we also store that
2017-07-15 19646, 2017
alastairp
we've not done this yet, but the plan is that we can go back and look at all this data
2017-07-15 19606, 2017
alastairp
and using what we have in musicbrainz as a base, provide a mapping, saying that "all of these messybrainz ids are actually this musicbrainz id"
2017-07-15 19616, 2017
alastairp
from there, you can do your normal recommendation
2017-07-15 19651, 2017
cjnolet_work
Ok. That's awesome!
2017-07-15 19652, 2017
cjnolet_work
Thanks so mcuh!
2017-07-15 19616, 2017
alastairp
(don't thank us yet, none of this matching has been done yet...)
2017-07-15 19620, 2017
alastairp
it'd make a good paper
2017-07-15 19623, 2017
alastairp
hint hint
2017-07-15 19632, 2017
cjnolet_work
hehe, yes!!!
2017-07-15 19654, 2017
cjnolet_work
So I've also been working to collect some data (really interested in spotify at the moment) and I hope to contribute that as well
2017-07-15 19627, 2017
cjnolet_work
Trying to get with professors right now to figure out the best way to incentivize students and others to help us out. I'm almost thinking as far as going into an ICO where I can give people coins to help us out
So far I've only gotten a small handful of people to contribute data and it's all Spotify data. nobody I know has last.fm accounts, doh!
2017-07-15 19652, 2017
alastairp
interesting
2017-07-15 19626, 2017
cjnolet_work
I'm really only interested in the final recommendations that come out of this. So I'm really more interested in just helping integrate the data into listenbrainz
2017-07-15 19627, 2017
cjnolet_work
It got me thinking though- if this was an ICO and a cryptocurrency was used to fund people giving data, it might be possible to create an automaton mechanical turk
So an ICO is similar to an IPO (stock offering) accept it's cryptocurrency instead of stock. Startups have been using ICOs to gain initial capital and get funding to do great things.
2017-07-15 19617, 2017
alastairp
http://digitool.library.mcgill.ca/R/ and search for "Evaluating the performance improvement of a music recommendation model by using user-centric features"
2017-07-15 19622, 2017
cjnolet_work
It basically creates a coin economy that is generated REAL capital. e.g. people buy coins for services, then coins are given out to render services. The exchanges of the coins create real fiat value
2017-07-15 19658, 2017
alastairp
work done by a colleague of mine, using last.fm data
2017-07-15 19614, 2017
cjnolet_work
Oh that's awesome! I'll print this out and give it a read.
2017-07-15 19622, 2017
alastairp
he said that he'd publish the data that he collected, though I'm not sure when that will be
2017-07-15 19636, 2017
alastairp
I'll definitely mention it to you when I know it's been done
2017-07-15 19642, 2017
alastairp
might be useful data for you
2017-07-15 19609, 2017
cjnolet_work
I'm more specifically interested in doing group reocmmendation
2017-07-15 19633, 2017
cjnolet_work
like being able to have multiple users with different taste profiles and being able to make playlists that all of the listeners woudl like. That's what the muse.fm app I'm building is for
2017-07-15 19601, 2017
cjnolet_work
the prototype is actually done, but I'm in the cold-start problem, lol
2017-07-15 19602, 2017
alastairp
that's easy, just play top40 :)
2017-07-15 19623, 2017
cjnolet_work
haha yeah, when the users don't have any tastes, I find the most familiar and popular tracks
2017-07-15 19646, 2017
cjnolet_work
I just don't want to release another streaming radio app and then get stuck in the same revenue problem as pandora, 8tracks, and spotify
