yeah, so - extracted features are a matrix - x axis is time and y axis is frequency
and each value is a 32 bit float (or 16 or 8 depending on how much data you think you need)
"You played this a very reasonable 422 times as is your right." i hear sarcasm.
lucifer: should I finish things and then add tests? or should just finish things, doc and then do a separate PR for fixing tests?
so, an alpha-rgb png is 32 bits per pixel (8 pixels per channel)
ruaok: tests in a separate PR is fine.
What in hell did you listen to 422 times lucifer ?
(also, maybe next year you can aim to listen to smething 666 times? just a suggestion.)
if you take the bits for a 32bit float and chop them up into 4 chunks of 8 bits, then you just use that as the colour value for a png. and now you have all of your audio features efficiently stored in a png
alastairp: I'm surprised the vocals are mostly understandable at every level for the Tankian one
ugh how insufferable attitude
CatQuest: oh spotify actuallly did you are in the top 1% listener of this artist
sarcastically i assume?
Last year it was not, but maybe this year it is :D
don't think so :) even has a link to tweet thanks to the artist
I still say we good-naturedly rickroll people
ruaok: the only playlist spotify made for wrapped is your top tracks!! looks like we have something better than them :D
s/only playlist/only personalized playlist
that sounds desperate...
My top is 43 times, but they said it with "very reasonable" too
huh, maybe they just say very reasonable for everyone then? i'd like it more if its sarcastic though.
huh, I listened to more music than 96% of the US listeners
94% for me.
reosarevok: I usually run then from inside the production-cron container. something like sudo -E -H -u musicbrainz carton exec -- ./admin/RunReports.pl DuplicateArtists
(you have to run the script with carton exec but also as the musicbrainz user)
overall, spotify wrapped looks great. the stats are great. top stats + moods + genres. if you replay wrapped, even different things show up some times. more playlists is the only thing i feel lacking in it.
as usual, it was flashy. I thought the 96% stat was cool. and your top genres was cool, but not surprising.
I think ours will have more depth, TBH
will be fun to see them side by side
i liked the top listener to the artist too.
spotify's wrapped has been completely a review of the user's listening. it does not have any music discovery. LB's similar users and top missed tracks so that definitely adds more detail to it.
*new music discovery
bitmap: I'll just run them all, since it seems several are missing and I can't be bothered checking them all one by one
this is generated by the similarity API call with neighbors = 1000 and dedup=all. then all tracks duplicate tacks (based on recording name and artist name) are removed. chopped down to 50 and uploaded. minimal processing.
v6lur joined the channel
ruaok: i see. interesting. but i haven't listened to the seed recording or the neighbours before so can't say anything about whether these are good results or not currently.
v6lur has quit
the seed recording is usually the first one in the list.
if you give me a seed MBID, I can generate the two lists for you.