I have trained different regression models to predict valence and arousal based on tags and acoustic feature of a dataset of songs.
The ground truth for valence and arousal had non limited values, but it generally went from 0 to 10.
Since in the first trainings i had different mse values for Valence and arousal (generally valence mse was 0.5 above arousal mse), i was wondering if using a standard scaler on the dependent variables (i already scaled the indipendent ones) before training could help these disparity.
But my question is: once i get the resulting mse values in the trained models with scaled ground truth, how do i effectively compare these values to the old ones? i need to unscale them someway, but i don't know how.
i scaled using sklearn.preprocessing.StandardScaler
