Training supervised call classifier - inspect misclassified calls

[imgross]

Hi,

when I train a supervised classifier to discriminate between two different call types I eventually get a confusion matrix, showing how many of the calls were correctly and how many very incorrectly classified.

Is there a way that I can look at the misclassified calls and see whether they share some similar features to get a feeling which calls are confusing the classifier?

Thank you, Isa

[DrCoffey]

There is no built it way, but if you debug the code (click on the dash next to line# 160 in TrainSupervisedClassifier_Callback.m) you can look at the variables in the workspace after running the training code.

On Tue, Aug 20, 2019 at 7:18 AM isa notifications@github.com wrote:

Hi,

when I train a supervised classifier to discriminate between two different call types I eventually get a confusion matrix, showing how many of the calls were correctly and how many very incorrectly classified.

Is there a way that I can look at the misclassified calls and see whether they share some similar features to get a feeling which calls are confusing the classifier?

Thank you, Isa

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/DrCoffey/DeepSqueak/issues/55?email_source=notifications&email_token=AJOFGE75NEJI2SAFDCCG5TTQFP4KNA5CNFSM4INXPLM2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4HGIKGSQ, or mute the thread https://github.com/notifications/unsubscribe-auth/AJOFGE6FOYH52ZXRCL77CDDQFP4KNANCNFSM4INXPLMQ .

-- Kevin Coffey, Ph.D.

Senior Fellow University of Washington

Harborview Medical Center 325 9th Ave Seattle, WA 98104 USA

mrcoffey [at] uw.edu mrcoffey@uw.edu (860) 874 5659

[ARTUR551]

I had a similar problem, I want to review the entire spectrogram to see which calls are missing from the detection results. I saw that some guys did something about this, you can review the hole file and edit the detection area (moving it, or making it bigger or smaller). https://github.com/UEFepilepsyAIVI/DeepSqueak

Is it possible to add that feature to the original DeepSqueak?

Thanks for DeepSqueak guys, it is great!

Arthur

[DrCoffey]

We are working with the Robert from the Screener team to obtain funding to improve DeepSqueak. This will be the first new feature added.

[ARTUR551]

That is great! Thanks!