You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
With CFG.NULL_VAL = 0.0, three metrics (MAE, RMSE, MAPE) are defaulted as masked ones. Maybe you should use Masked XXX instead of XXX in experimental results?
The text was updated successfully, but these errors were encountered:
Hi thanks for you suggestions.
The “0” is the abnormal value for traffic datasets. Thus, the “MAE” we wanted is indeed the masked ones, since calculating errors on abnormal values and forcing models to fit these abnormal data does not make any sense. Almost all the previous works followed this setting and reported these metrics without the “masked”.
Moreover, the default null value of these metrics is nan, instead of 0.0, since in many datasets the “0” is not abnormal values.
With
CFG.NULL_VAL = 0.0
, three metrics (MAE, RMSE, MAPE) are defaulted as masked ones. Maybe you should use Masked XXX instead of XXX in experimental results?The text was updated successfully, but these errors were encountered: