EON Tuner: does not run for 2 of 3 classes in confusion matrix

When I run the EON Tuner on audio samples, it does not run for 2 out of 3 classes in the confusion matrix. It reports 100% accuracy for the 1 class it successfully runs for, but shows ‘-’ signs for the others.

Project ID:

Context/Use case:
I have 3.5 hours of audio clips that should be classified into 3 classes (e.g. “Apple”, “Orange”, “Pear”). The audio clips are from 200ms to 5 seconds in length. I used the default settings for almost everything across the project so I basically haven’t changed anything (e.g. I haven’t written any code). This is my first time trying out the tool, so it could be me doing something wrong.

Here is what the confusion matrix looks like after running EON Tuner:

Update: I ran it 2 times, with the same result.

Hi @sunflower55,

Have you changed the EON Tuner Search Space parameters? The window size parameter was set to 4000ms. As only your Unknown samples have duration longer than this value, it was the only class being classified.

I’ve changed it to the following values and it now goes through:

Also note that your dataset is very unbalanced so the NN will probably have troubles learning the other classes.


1 Like

To answer your question, no because I either assumed the defaults were fine or didn’t notice the “Space” tab in the UI (it’s quite light gray). For window size being 4000ms, I mistakenly assumed it was the upper limit of inference time that I would accept and that it mostly didn’t matter. I’ll have to educate myself on this more now, thank you Aurel.

1 Like