I am wanting to run a series of models in the automatic time series package, AutoTS in python. One of the models I am trying to run as one of the contenders is a naive model. However, whenever I try to run a naive model the autoTS package always runs it with transformations:
I use the following code:
modellist=["LastValueNaive"]
metric_weighting = { 'smape_weighting': 0, 'mae_weighting': 100, 'mlvb_weighting':0, 'rmse_weighting': 0, 'made_weighting': 0, 'mage_weighting': 0, 'mle_weighting': 0, 'imle_weighting': 0, 'spl_weighting': 0, 'containment_weighting': 0, 'contour_weighting': 0, 'runtime_weighting': 0, } forecast_length=1
model = AutoTS(forecast_length=forecast_length+3,prediction_interval=0.95, model_list=modellist, frequency="infer", ensemble="horizontal", no_negatives=True, max_generations=1,num_validations=0,models_to_validate=20,transformer_list={},preclean={"fillna":'rolling_mean', "transformations": {}, "transformation_params": {}},metric_weighting=metric_weighting)
model.fit(data,weights=weight,future_regressor=regressortrain)
Despite, omitting transformations in all possible places, such as in the pre-clean and by setting transformation_list to an empty set, the model still runs the naive forecast with transformations that materially influence the prediction; I am wanting to run a pure naive forecast through autoTS that is nothing more than the last value. How do I do this?
In all cases the prediction coming from autoTS, currently, is much worse than a pure naive forecast run outside of this package and I suspect this is due to the transformations it runs.
