#metabrainz

ok, big standard normalization.

*bog

that works well for linear data, but this data is not linear. and I think that is why the graph isn't very dynamic.

ok

You have any suggestions

but I think for now you should move on to other things. I'll have a play later today or tomorrow -- this is a problem I need to solve for another project as well. one solution would be good for both.

Ok that's really helpful

have you tried making the text bigger for a bigger bubble? and on two lines?

I have tried but there seems to be no easy way to do it

I can try looking up a bit more though

I did more work on the colors

And for the color scheme is it good or needs some changes

I did talk to aerozol: and he suggests we use a diverging pallete with one color from top track and another from top album

do you know the size of the bubble in pixels?

how do you specify the font for the bubbles?

I slapped together a quick script that fetches similar artists and prints the score, its log2, log10, and sqrt().

sqrt looks *great* if you ask me.

you should normalize sqrt(score), rather than just score and I think things will look better.

It works great on larger numbers

For some artists the number are much smaller

and it is not pretty in that case

For example some artists have 3 digit score

here is the data for blümchen, a much less played artist on LB.

the sqrt has a nice spread. let me normalize it.

ok

blümchen, normalized.

portishead, normalized.

So basically what I am doing is just normalization right now

Right ?

both of these have a decent spread between 0 - 1.

correct. so normalize the sqrt() and everything should look better.

Ok I will try that

So I keep the scaling function as it is and use sqrt inside it instead of just score

be careful. you need to calculate the maximum value.

but the maximum value needs to be the sqrt(maxium).

for working out the range of the numbers, don't do sqrt() more than once. sqrt() is *very* expensive.

So is it okay if I share the function and you can see what values I should in

I am a bit confused

let me post my code.

Sure

Ok it's much clearer now thanks

great. try it. I'll be around for a bit and we can see how it looks.

Sure I will let you know in a bit

k

mayhem: And we should take the inverse of the sqrt(score) / hi right?

So hi / sqrt(score) in this case as the greater score should be closer

yeah. just subtract from 1.0.

ok

mayhem: https://w7rshq-3000.csb.app/

You can see the changes now

feels much better, no?

and the shades also dynamic now, or are there only three colors?

No only three colors

would be nice to see if the improved scores would make the colors look better.

I guess this method is a bit faster as well

But the only thing is sometimes it becomes too large and exceeds the area

But I decreased the multiplier a bit

I am multiplying the computed values by 500

and sometimes it is too small.

Yeah that too

yeah, we will need to revisit that and make it more dynamic.

I feel the data is a bit all over the place and causing problems

100 similar artists is all cut off.

the data is all over that place and that is not going to improve.

and the data problems we just took care of.

what we have now is an overall scaling problem.

True

and that is related to how they graph is drawn and doesn't depend on the data.

Yeah

Should the colors be based on the scores as well

Or should there be groups like it is now

the normalized scores, yes.

Ok and use two colors

This is the mockup aerozol: made

yeah, two colors is right.

Ok I will try that as well but couldn't find an easy way to generate a divergent color pallete

There are some in built with nivo though

But in limited number

you mean a gradient?

If you see the divergent option on this

why do you need a pallete generator?

ah, i see.

you're making it more complicated than it needs to be.

here is the algorithm you can use:

So wanted to get your thoughts as well

hmm. ok, I guess we should really ask him.

Right now I use something similar to this, https://stackoverflow.com/questions/40619476/ja...

for me, at first glance, I would treat the gradient as a value from 0 - > 1.0

if you want to pick the mid-point, ask for the color at .5

Yeah right now I generate shades in a order of 0.2

so 1 -> 0 in 5

and then for each red, green, blue do: ((color2.red - color1.red) * index) + color1.red

where index for the mid-point is .5

then you have a smooth gradient between the two colors. and I think we should implement that and see how it looks.

ok so color1 and 2 are the nodes right ?

or the color of the nodes

those are the two given colors from which we calculate all the others.

ok I see

the root node will be color1. the outermost node will be color2

all nodes inbetween will be a color inbetween

Understood

I believe that is a good way to go

👍

the results are looking good!

I will try that out and see how it goes

before you finish work for the day, can you post an updated link for me?

I'm off to bed now.

Sure will do

Appreciate your help

And good night

np, thx

mayhem: arsh: sorry, just catching up

Was there a specific question for me? mayhem I tagged you in this a few days ago, here’s the mockup, if you have questions or feedback go for it https://usercontent.irccloud-cdn.com/file/KFkTW...

That was the first pass, I didn’t get any feedback

If it slowly changes gradients every time you click it should look very nice and dynamic. But if it’s too hard to implement then I can come up with something else

yeah, sorry slipped off my plate today. will look in the morning

does the palette change a tiny bit for each click? making a revolution around the hue circle?

This example is based on an idea of arsh’s to use huesound to grab the colours from the two covers pictured. But if huesound doesn’t work/is too intensive then we can randomise it. Or go around the hue circle - but random could be fun, since the colours used aren’t overwhelming (they fade out to 20% on the outside, bg is 20% as well)

Chat in the morning :)

aerozol: have you seen the latest version? https://w7rshq-3000.csb.app/

Yeah! Looking good imo!

ok good. what sorts of things do you see that need to be fixed/implemented still?

the text needs work, for sure.

the overall size of it need to scale to take up 85% of the window

A few things (comparing it to the mockup), do you mean colours specifically?

I'm generally in favor of coming to a stopping point on the graph as soon as possible, since there are many other things to do still. we can come back later to do more tweaking if there is time.

overall -- I get the idea you're shooting for for color. I think I will guide arsh to implement the walk around the hue circle for an initial version and then move on.

Right, I don’t know your schedule. But without the gradient and the colours converging towards them it’s not like the mockup

understood -- he is working on the gradient now.

👌

what do you mean by colors converging?

or do you mean background gradient.

I guess, yes, I have questions about the colors. :)

in your mock-up the graph is more or less in circles, which is not how things are shaking out in the demo.

which means assigning 4 colors is somewhat arbitrary.

The circles are hue A (at %100) fading towards hue B (at %20) as it goes towards the outside. The background gradient is hue A and hue B both at 20%

I've asked arsh to work on coloring the bubbles based on the distance between the innermost circle and the outermost circle.

That’s ideal