Fluid.plotter labels ( pure data / linux )

Getting Started and Wayfinding
1

Hello,

in order to test flucoma
I adapted a max patch in pure data,
that slices and analyzes a sound, then plot in
2D it seems to be working , then I tried
to add labels, and failed. I must be doing
something wrong, I spent some time trying to figure out,
here is the patch attached, if anybody can help
me find out.

Cheers

fluidCorpus2D.pd (8.1 KB)

2

ok the problem is triple:

  • Pd transforming float to strings. We need to change our examples because 0.00000 seems to not be recognised as a labelā€¦ in your code I have added a [makefilename %d] object in the label making in [pd mfcc analysis]

  • if you use the nightly builds (you should for the fixes) then you will see I have made fluid.plotter in C - faster but I didnā€™t include 8 colours in my array so 8 classes freaks out - if you change numclusters to 6 youā€™re fine (for now, Iā€™ll add another 10 colours in the array and deal with the size cap too in a more elegant way) 16 classes should be enough to plot no?

so this patch works for 6 clusters. Iā€™ll correct the C code to allow for 16 clusters in [fluid.plotter] later today or tomorrow.

p
fluidCorpus2D.pd (8.2 KB)

3

Thank you !

I did notice the floats as indexes in my datasets and added the ā€œmakefilename %dā€ but didnā€™t think about reducing the cluster number.

I noticed when using ā€œhighlightā€ message that after while the selection of points in the plotter is slowing down.

4

yes but that is Pdā€™s graphic APIā€™s fault. It is very old. like 30 years old. So the more you add points, the slower it gets. That community is in discussion to change their graphic API (which will mean a painful recode from us but for a good cause)

5

So Iā€™ve pushed a fix - it is compiling on the ā€˜nightliesā€™

Iā€™ve used 16 equidistant colours from CET06 (CET Perceptually Uniform Colour Maps) as we did in other places. I could have used the 10 we use in SuperCollider but that is only 10 classes (https://github.com/d3/d3-scale-chromatic/blob/main/src/categorical/category10.js) although it might be better for cohesion.

if you can live with ā€œonlyā€ 10 categories, let me know and Iā€™ll change them for a more unified fluid-world :smiley:

1 Like
6

actually as geek is a geek - I couldnā€™t resist to compare. @tedmoore do you have a preference?

Screenshot 2024-02-26 at 11.01.13

Screenshot 2024-02-26 at 11.03.21

if we feel 10 is enough I think I might prefer the SC colour schemeā€¦

comments welcome, change is trivial, unified world is my favourite option as usual

7

ok 10 will have to be enough - this is so cute:

Screenshot 2024-02-26 at 11.08.34
Screenshot 2024-02-26 at 11.08.34648Ɨ634 141 KB

8

I donā€™t really have a preference. If anything Iā€™ll observe that if the black circle around the dot needs to be there, then the lighter colors are better perhaps, they help differentiate, but I donā€™t know if thatā€™s the 10 or 16 category scheme.

But as you say, itā€™s not too important.

9

I donā€™t have an opinion on how many classes is enough, Iā€™m slowly discovering the Flucoma library since last november, I like the Flucoma pd documentation a lot by the way, the patches are instructive both about the library and pure data itself

2 Likes
10

1.0.7 is SUPER FAST now that I found a hack around the (curse word) C+TCL interface memory issues.

try it in this thread:

feedback welcome. Iā€™m offline until the 2nd but hey, road test it please!

11

Hi, Iā€™m testing this patch in Pure Data and I was wondering how it would be possible to playback the analyzed files in stereo.

thanks

12

Hello, I will take a look!

13

fluidCorpus2D.pd (8.3 KB)

This will play a stereo file ( though the analysis will happen on the left channel only )

1 Like
14

Thanks) work awesomeā€¦