Issue

This Content is from Stack Overflow. Question asked by Funzo

I want to use kats to forecast sales of product we will sell.
One of the key indicators of how mcuh we will sell is how much we spend on advertising leading up to a given day.
We also know this for the coming weeks.
So we want to provide that this at prediction time to improve prediction accuracy.

I am following this tutorial https://github.com/facebookresearch/Kats/blob/main/tutorials/kats_201_forecasting.ipynb.

I don’t see any way of giving the model this information.
Is this something that is not possible?

Solution

Thanks for the question! For the current available models in Kats, ‘prophet’ supports covariates. User can specify the name of covariates in the ProphetParams and feed a multivariate TimeSeriesData (containing the corresponding covariates) into ProphetModel. To generate forecasts, the future covariate information should be passed as a pd.Dataframe object like
pm.predict(steps = 10, future=pd.DataFrame({'ds':..., 'covariate':...})).

This Question was asked in StackOverflow by Funzo and Answered by YBK It is licensed under the terms of CC BY-SA 2.5. - CC BY-SA 3.0. - CC BY-SA 4.0.