Question/Issue:
What are some reasons for the EON Tuner to show a lower accuracy than the classifier model accuracy?
Project ID:
195638
Context/Use case:
I’ve been running my model multiple times with different settings, getting what seem to be good results from the NN Classifier (97-100%). However, when I run the EON Tuner and sort for accuracy, the best result is more than 15 % lower. My dataset is somewhat unbalanced, with the smallest class having samples that last 1m8s and the largest 1m29s. Is this a sign of overfitting or something else? If the model had been deployed, would the accuracy on the device be that one from the EON Tuner or from the NN Classifier?