Processing files stuck at 50%

bthj

Hi,

When selecting to create a new sound map from a folder with forty thousand files, with the default settings, the process reaches 50%, stays there for a while with a separate process using all CPU cores, then that seems to complete and the main "AudioStellar" process is left consuming around 64% of one CPU core, and the GUI progress bar is still stuck on 50%, after around three hours - so it seems stuck, but maybe I should give it more time?

This is on an Apple M1 Pro chip, macOS 13.4.1 (22F82), running AudioStellar v1.3-beta2

The sample set I'm trying to process can be (temporarily) downloaded from here (8GB):
https://uio-my.sharepoint.com/:u:/r/personal/bthj_uio_no/Documents/Attachments/evoruns1-last-iterations-wav-render.zip?csf=1&web=1&e=rb2Wj8

bthj

I've had better luck applying UMAP to this dataset, instead of t-SNE.

After a successful process, if I choose to create a new sound map, tweaking some settings, then the app eventually crashes, and sometimes the whole OS (probably due to memory consumption). Restarting AudioStellar before choosing to run a new map works better.

I'm wondering if it would make any difference to compile directly for the Apple silicon I'm using, instead of using the Intel code through Rosetta, and if I could give that a try: Are the compilation instructions here up to date?:
https://gitlab.com/ayrsd/audiostellar#how-to-compile

would I have to go through step #6, the data-analysis instructions?

bthj

One update: UMAP with canberra as the similarity measure hasn't finished after twelve hours, while the same settings with either cosine or euclidean complete in a few hours.

macramole

it sounds like you are running out of memory , is the sound map you achived not good enough ? cosine distance should be ok, no need to change that. UMAP works better than t-SNE, we are using a different t-sne implementation in the next release

bthj

Yes, that seems to be the issue. The sound maps I've achieved with UMAP and some of the similarity measures, for example braycurtis, are good and I'm having great fun exploring!

With a large set like this, the points overlap quite a bit, so I wonder if it would be viable to have to option to spread them out evenly, for example with a slider? - maybe that's a topic for another thread.

macramole

That would be nice. Do you know you can manually move sounds dragging with secondary mouse button ?

bthj

Thanks for the tip. I've tried that and it's definitely useful.

macramole

it is also useful for sequences, so you can change the rhythm by moving the dots around.

bthj

That's a really innovative approach to generate rhythms.