#metabrainz

[listenbrainz-server] mayhem opened pull request #1293 (add-mbid-mapping-to-labs-api…add-recording-search-to-labs-api): Add user facing recording search https://github.com/metabrainz/listenbrainz-serv...

phew. these two PRs have been on my mind for 2 months now. finally out!

!m ruaok

You're doing good work, ruaok!

ruaok: just as a headsup, LB is currently in an undeployable state, yvanzo's last patch requires consul-template 0.18, which is coming in #1237, I hope that _lucifer and I can deploy this tomorrow

i am available currently also if we want to do it today :)

ok, good to know. I don't think my PRs have a snowball's chance in texas to get approved before you two fix the LB codebase, so....

_lucifer: I don't have any time today, but you can have a look if you want

alastairp: is all the code ready to go, so I can help _lucifer with the release? I have some time left in my day.

no, I don't think it's ready

ok, then I'll leave you to it.

I rebased 1237, but I didn't realise that your 1282 was a PR onto that branch, so now we've got major conflicts between the 2 PRs

_lucifer: if you're idle for a moment, we can chat about the spark work I am proposing for next week

it's going to take some git-fu to unwind it

ruaok: sure

doh. I was really hoping to make it easier by making a branch off a branch.

not a good idea or does it need more highlighting in the docs.

BTW: I've done that same thing again, with my two PRs for today.

"not a good idea or does it need more highlighting in the docs." +?

yeah, I marked my PR as WIP (at least in the title), because the initial changes that I did were just experiments, so I had planned to rebase it until I had a good solution

perhaps we could be clearer about this. I think it was right that you made your one base off of mine, but perhaps we should have merged it sooner

let me leave a comment in the future to make this more clear.

if i understand correctly, we want to rebase 1282 over 1237?

merge master into my branch does what I mostly need.

but that doesn't answer your question.

:)

_lucifer: 1282 was supposed to be merged into 1237 before reviewing 1237 and then merging both into master at the same time.

_lucifer: https://www.dropbox.com/scl/fi/tcffmk725ylfx3me...

the user similarity feature is what I think you and I should tackle next week

the core of the feature is explained in "The actual calculation of the similarities is a statistical problem. Each user can be represented as a vector where the nth element of the vector is the number of times the user has listened to the nth artist. Once each user is represented as a vector, Spark provides an API to calculate the Correlation matrix for a list of vectors where matrix[i][j] = the correlation between the ith and jth

and that is clearly the right approach for this problem.

lol

do we want to allow users to see similiarity with another user or just show similiar users to them?

i would prefer A

so that any user can see the similar user list for any other user?

yes

spark supports B out of the box

we should allow that.

if i'm logged in and I go to ruaok's user page, i should see our "compatibility score"

is that a function of spark or just showing the right data?

then we can show follower's compatibility and such along with most compatible users or something.

do we need to calculate different data for the two approaches? I dont see that....

atj: No problem, can you please explain how you did resolve the issue so we can add it to a troubleshooting section for other macOS users?

ruaok: if you're just showing the user other similar user's, you can get away with just getting the similarity for the top X users.

but for A, you need the entire correlation matrix.

ah, its the same algorithm, but the number of results we keep differs. is that right?

yes, A would allow seeing the similarity between any two users but B would only show between the top X similiar ones

ok, that makes sense.

but, that matrix will be VERY sparse.

yeah.

and 90% of the users will relate to each other poorly which is useless info.

which makes me think we should have a cut-off where don't keep the users. it doesn't scale otherwise.

so it really only matters on how we want to show the data to users.

exactly.

if it's just compatability scores, i assume we can just do scales of 1 to 5 and get away with it

[listenbrainz-server] shivam-kapila opened pull request #1294 (master…ws-rec_msid-fix): LB-821: Fix the position of recording_msid in WS messages https://github.com/metabrainz/listenbrainz-serv...

yeah, something like that.

so just store 2-5 for ranges, if the correlation is too low, just don't store it.

I think a scale of 10 is better, but...

"ohh, you're a perfect 10! 😍"

makes sense

sounds good.

also one correction, i had confused the B part with something else.

what spark supports out of the box is recommending a set of users for a given recording.

ok, I need to get moving now.

I'll keep working on this doc tomorrow, now that my PRs are out.

We can use phrase like:

1. oh you share some music taste. Nice

2. You look like music buddies.

3. Damn. Are you sure you guys are not jamming twins!

[musicbrainz-server] atj closed pull request #1932 (master…MBS-10312-fix): MBS-10312: Fix tracking parameter removal in Firefox 44 https://github.com/metabrainz/musicbrainz-serve...

yvanzo: wasn't sure where to add it https://github.com/atj/musicbrainz-docker/commi...

atj: you can edit the branch something points to, it might have been just fine :)

Anyway, did you manage to test in the end? if not, we can throw the patch on http://test.musicbrainz.org/ when you're ready :)

reosarevok: it's turned into a bit of a nightmare :)

I was previously tracking whether we made any changes to the url and if not returning the original string

Oh no :(

however now we return url.href unconditionally, which has revealed that the URL object does weird things to unicode URLs etc.

so it's actually revealed other bugs

Eek, a fractalbug! The worst kind.

Project bookbrainz-site build #3657: passed in 6 min 5 sec: https://travis-ci.org/bookbrainz/bookbrainz-sit...

e.g. https://baike.baidu.com/item/啊呀啦嗦 and https://baike.baidu.com/item/%E5%95%8A%E5%91%80... are actually the same

url.href returns the latter

..so i t was punycode after all :P

heh. https://baike.baidu.com/item/%E5%95%8A%E5%91%80... automatically turns into https://baike.baidu.com/item/啊呀啦嗦 in the urlbox of the browser. but if I try copy pasting it from there again it's https://baike.baidu.com/item/%E5%95%8A%E5%91%80...

Ugh.

Project bookbrainz-site build #3660: passed in 4 min 23 sec: https://travis-ci.org/bookbrainz/bookbrainz-sit...

Project bookbrainz-site build #3661: passed in 12 min 58 sec: https://travis-ci.org/bookbrainz/bookbrainz-sit...

CatQuest: Set `browser.urlbar.decodeURLsOnCopy` to `true`

Oh wait, it seem that pref doesn't exist in old versions: https://bugzilla.mozilla.org/show_bug.cgi?id=13...

Workarounds: https://bugzilla.mozilla.org/show_bug.cgi?id=13...

reosarevok: I think the options are a) back out the change or b) fix some of the test URLs

e.g. https://www.whosampled.com/Pet-Shop-Boys/Can-Yo... isn't a valid URL

and it doesn't work, the URL is actually https://www.whosampled.com/Pet-Shop-Boys/Can-Yo...

we should tell https://www.whosampled.com/about/ to use our updated logo :D

[bookbrainz-site] MonkeyDo opened pull request #556 (master…snyk-fix-b36b5c206c21b7aa93f63d90135b1314): [Snyk] Security upgrade lodash from 4.17.20 to 4.17.21 https://github.com/bookbrainz/bookbrainz-site/p...

Project bookbrainz-site build #3664: passed in 4 min 33 sec: https://travis-ci.org/bookbrainz/bookbrainz-sit...