#metabrainz

right!

i didn't wnat to sound liek a complainer!

and i realise that the fonts being to large is a thing with *me*

Better than saying ' the home page doesn't look good' once it's all done...

there are probably scripts or user prefs that cna be made

oh it looks "good"

Hmm, there's a part in accessibility about accompanying images with text whenever possible

but uh. it's very distracting right now :D

yes!

bot alt and visible text, please

But for some things like 'expand section' an icon with alt text is probably fine? As it's expected?

What do we have icons for that bothers you

yep, that's fine

and x's for removing something and +'s for adding

those make sense

I've even gotten used to the pencil icon for editing (although a little button that said "edit" wouldn't be hated :P)

Bit late, but: I agree the BB gradient colors need to be adjusted, the original brown look meh as mentioned but the black+orange makes me think of halloween more than books :p

I have explisit suggestions for some pages, once we get to thme i'd love to explain

it's rather dramatic

why not opt for a more golden brown so it looks golden-y

or isn't the actual bb brown a bit redder thna that? reddish brown woudl look good

Well, it's great that we're talking about this, because I don't even know if we want to have the other projects move towards MB? For instance: using gradients

I still want the user, data and admin menues to the right though

I've even made a script to do this s it 's all the way to the right

i *need* them there lol

personally I could go witouth gradients. maybe only used in spesific places?

i liek the idea someone had of having a slight colour diffence based on wich entity page ou where on

akshaaatt: you should be able to access that doc now too, it wasn't open to @meb team

lord knows bb could do with entity-page differenciation

Thanks alastairp

what I *do* like bb does: buttons are thematically coloured, so a remove button wwil lbe red, and add button wil lbe green, and edit one will be orange

Ooh, will add to my notes. Buttons is a no-brainer for aligning across sites tbh. I don't know any reason not to

some more polish to the system maybe, but the idea is sound

well mbhas A LOT of buttons that does differnt things

(even buttons that imho shoudl be links...)

aerozol: how about a : don't use fancy code to do the smae thing simple code can do (eg use simple css isntead of complicate css)

if something van be accomplished with a simple div don't use inpage csss and junk

agreed, but the design guidelines don't venture into code at all at this stage

alright

I've just said 'use the design library, or copy other projects'

so my concern has mainly been consistency

yea. there is some issues

afaik lb, cb and bb are all pretty similar

mb has more legacy

p.s. a use of the gradient that I think would be really nice is just a small strip above the footer, for instance

that's where it gets really tricky - are we making everything look like lb, cb and bb?

oh yes, a 1-3pixel high thing like, sure i liek that

honestly I would like more mb-fication of the former rather than making mb like the others

(with some exeptions, ie colourful buttons)

for accessability: ther shoudl be an ability to make fonts larger

or, smaller

on buttons: MB has a lot, but I think they can probably be grouped into a few types. 'link' 'edit' 'navigation' etc

yep

well links are edits

I'd like a colour for edit history

oh yeah that didn't make sense hah

turquoise..

that'd be neat

To make it stand out?

yes

(also it's one of my favourite colours :D)

Well, we better make everything turquoise

ok so. i have some uscer css things on mb. and one of those is greating a grey "border" like background on mb (becase it used ot have that) and i miss this on the other sites, becase the slight border around pages lets me concentrate on contence of the pages

nah, we an have orange-turquoise as thematic colour for GameBrainz or something

you better believe VideoBrainz and GameBrainz are going ot be things

is release_group.last_updated bumped when an associated record is changed? E.g. when a new cover_art is added via one of its child releases?

bumped how?

oh yea

🤔

ac it's dinner tim i can't think. must eat

bitmap: ^ (q3lont's query)

CatQuest: Yea.

Especially when data scraping has become so stupid easy now a days

q3lont: no only when that row is updated

btw listenbrainz-dev: i get a type error on this page https://listenbrainz.org/user/outsidecontext/ but not on other "listens" pages

thanks

I have noted your feedback down btw CatQuest

cheers. thnaks for listening ot this old wet sock <3

CatQuest: maybe some temporary issue. can you reload? it loads fine for me.

lucifer: have reload many times. reos page and my own works fine

oh weird.

there is a lot of "TypeError: t is undefined" errors in console

lemme see

monkey: ^

right off to food and no more screen for me today. i am tired

see yas tomorrow

thinking of moving my pet project to a VPS so that I can run musicbrainz-docker alongside. Updating production db data from my local MB setup is no fun :/

mayhem: playlists with new confidence function in. it chose rank 21. rmse is 22. can probably go for rank 50-100

Let's try it, why not?

playlist looks good currently as well. fwiw

the top artist is actually better now, similar artist one still decent but i like the previous one better.

[bookbrainz-site] 14dependabot[bot] opened pull request #839 (03master…dependabot/npm_and_yarn/redux-4.2.0): chore(deps): bump redux from 3.7.2 to 4.2.0 https://github.com/metabrainz/bookbrainz-site/p...

[bookbrainz-site] 14dependabot[bot] closed pull request #825 (03master…dependabot/npm_and_yarn/redux-4.1.2): chore(deps): bump redux from 3.7.2 to 4.1.2 https://github.com/metabrainz/bookbrainz-site/p...

Params: {'ranks': [5, 2, 10, 15, 20, 11, 12, 13, 14, 16, 17, 18, 19, 21, 25, 50, 100, 75], 'lambdas': [0.1, 10.0], 'iterations': [10], 'alphas': [3.0]}

:)

lol, ok.

this similar artist playlist is looking quite interesting. very plausible, but clearly treading on different ground than before.

cautiously optimistic about the new transfer function

curious to see if this will cause a shift in the ranks. it might very well do that and i think that would be a very good indication and give some confidence to the transfer function. but we'll have to see how that bears out

the similar artist playlist also recommended a couple of songs i hate but spotify still recommends me constantly so we are definitely inching closer to being par with them at least :)

those were towards the bottom though last 10ish.

yes, I feel your play history PR is a good step in the right direction, since I'm starting to listen a lot of songs I've listened to already.

its OOMs currently sadly. otherwise it'd be on test by now.

and also hated tracks need to get removed from candidate sets.

i'll debug it tommorrow to see why it does so.

oy. dreaded OOM

feels like a huge sigh of relief that things are getting on track.

indeed.

I still dont quite grok training numbers, but ok, rank seems to be responding.

and perhaps lambda will respond more.

i think we should add K-folds cross testing. training time'll increase but randomness in results will decrease probably.

unsure about alpha now -- my transfer function kinda already does that.

got a link where I can read up on k-folds? that term is new to me

basically instead of doing 1 split, do K splits. and train for each split. take the average rmse of each training. choose the model with the best average.

yes, please. the current training system seems overly simplistic.

it this something that spark makes relatively easy?

yes spark has apis for that so i think it should be on the easier side. but i haven't looked into deep enough to be sure but optimistic.

Of interest: I remember an article about Spotify that said the song it plays after what you picked is one that you would expect, a obvious pick. But that the one after that the user wants something that they like, but wouldn't necessarily have picked themselves

Peppering something into your tasty broth!

one thing though i like that. the LB playlists are quite close to what i put on daily on spotify anyways.

and this is without collecting any fine grained data or demographic etc spotify does,

lucifer: YESS, that is exactly the sweet spot where we should be. and now with more troi massaging we can start making actual playlists.

which of course gets into a whole new set of academic papers...

aerozol: yes, that is exactly what my last message alluding to.

the daily jams should most certainly start on a top track from the last month and then go to a high rated track that you haven't played in a month. then something new to you.

indeed, makes sense!

and I seriously hope that this is where troi begins to shine. the scripts are super easy to write and everyone can jump in and make their own version with their own spin. then we accept PRs.

sounds good. time for a for troi competition soon :D

invite people to make the best playlist algorithms they can think of.

I think it might be time to go public with this soon.

once alastairp gets the new datasets out on the FTP server, then I think we're ready.

we'll be a the "we've made an open source music recommendation engine that is okayish and doesn't track the living snot out of you"

point that I've been lusting for for years

lucifer: I can't wait to see what collaborative filtering does once you play down all the cool things it thinks you should listen to. what happens next?

yes makes sense

crap? discovery further afield?

not really sure. i think it'll keep recommening the same stuff.

I do like the "not track you to death" aspect you just raised. perfect bit of marketing.

it can't recommend the same stuff. we'll remove the "same stuff" from the candidate sets.

then what?

we'll need to do something else for new tracks to be recommended but there's always a limit.

yes, CF is not good at that.