Closed antoinecarme closed 5 years ago
Easy to implement,. A generic scikit-learn model can already be used (SVR models are OK).
A new package dependency
pip install xgboost
create anew branch pyaf_xgboost
Time series models based on XGBoost regressors has been added. They are not activated by default.
Need to activate these using something like :
lEngine.mOptions.set_active_autoregressions(['AR', 'XGB']);
Add XGBX models (past of the signal + past of the exogenous variables)
Transformed_Signal = Trend + Periodic + XGBoostRegressor(target = PeriodicResidue, input = PeriodicResidue_Lags + Exogenous_Lags)
Need to evaluate models of the type :
Transformed_Signal = Trend + Periodic + XGBoostRegressor(target = PeriodicResidue, input = PeriodicResidue_Lags)
Of course, this is done inside the competition (all possible combinations of transformations, trends and periodics are tested).