search

redouane-dziri / deep-music-classification

Classify music into genres using GLCM on mel-maps and CNNs
0 stars 0 forks source link

stacked_final model performance issue #19

Closed arnaudstiegler closed 4 years ago

arnaudstiegler commented 4 years ago

Just noticed that the stacked_final model has 36% accuracy on validation set in my evaluation notebook but the last model trained on the notebook in improvement/training_stacked.ipynb is way higher. I think this might be due to having the wrong stacked_final model save, but there could potentially be an issue with it. Won't be able to deal with that until Tuesday

redouane-dziri commented 4 years ago

I am re-running training and saving the model :) hopefully you won't have to deal with it

redouane-dziri commented 4 years ago

Figured out where the discrepancy comes from! The stacked model requires normalized input, so I just added a normalization operation before feeding the test data to predict on and now the accuracy is pretty much the same as the concatenated model!

redouane-dziri commented 4 years ago

You can close this once you've seen my reply if you have no questions :)