Using accuracy as metric for regression

[Liatokha-Ilya]

Thank you for sharing your work.

Can you elaborate, why are you using accuracy metric in regression task? You are predicting a vector of latent factors, which supposed to be a regression task.

https://github.com/andrebola/icassp2021/blob/bc551a5453f17728dacafdc74971337d4ee92e44/autotagging_vgg.py#L314

You also have a callback class in your code (which is not included in model compiling) and it uses ROC-AUC and PR-AUC classification metrics.

https://github.com/andrebola/icassp2021/blob/bc551a5453f17728dacafdc74971337d4ee92e44/autotagging_vgg.py#L183-L208

Can you explain why classification metrics are used for predicting latent factors?

[andrebola]

Hi @Liatokha-Ilya, thanks for your interest in our paper. When training the model to predict latent factors we use Mean Squared Error. Once the model is trained we use it to generate the latent factors for the new songs and use this to recommend songs to the playlists. The accuracy metrics are used only to evaluate the predictions for the recommendations.

All the information you can find in the paper. I hope that helps!