I am building time series models in Python using the statsmodels library.
Are there Python resources I can use to build ensembles of such models?
(My Google searches suggest that the answer is no.)
Contents
hide
Best Answer
Yeah there are a few implementations but nothing that looks highly maintained.
- https://github.com/carloschavez9/ensemble-time-series
- https://github.com/dushyantkhosla/ts-forecasting-ensemble
But if you are already using statmodels then you can make one yourself in short time. Typically you will want ets, arima, seasonal naive, naive, some decomposition method like stl and that will get you started. If you want to get fancy then you can stack these by using a holdout and training a model on top using the holdout forecasts vs actual to assign weights to each model.