I have a project (154) with MFE + Classifier doing keyword detection. Recently I’ve noticed that after playing with the model for some time (changing topology, retraining, etc.) at some point the Data Augmentation options disappear. At that point the only way to get them back is to blow away the entire model and start from scratch, at which point they reappear, but is obviously quite time consuming. I am going to rebuild 154 right now so if you look at it don’t be surprised if DA is present!
I should add that re-adding the classifier by itself does nothing. I have to remove the MFE block for the DA options to reset.
EDIT: It seems like this only happens when I “retrain model”
The project 154 belongs to another user, can you double check your project ID? It’s also displayed in the URL.
Regarding your issue, are you using the Keras Expert mode by chance?
Sorry, 164! And no, not currently using expert mode although I may have spent some time in expert mode in the past. The current impulse is fresh and has never been edited in expert mode.
I can see on the 137713 project that you are using the expert mode which does not allow (or at least not with one click of a button) data augmentation.
Would you be able to reproduce your issue in a new project so I can have a deeper look please?
I spent some more time investigating, I can reproduce on a fork of your project but not on some other projects. What seems to be different from the two projects is the number of filters for the DSP (thus the reshape layer of the NN).