Future values for multivariate predictions #179
-
Hi everybody, I have a dataset with 20+ features and trying to predict one of them 6 months into the future. Some of these features have known values for these 6 months to the future, how can I help the training/forecasting with these values. I have tried using Thank you very much. K. |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 3 replies
-
so what create_regressor does it is shift values that aren't known from the future into the future so they work as a |
Beta Was this translation helpful? Give feedback.
Firstly, you should know that
future_regressor
is optional. Internally, for machine learning models (ie MultivariateRegression with regression models like LightGBM and ExtraTrees) it will automatically create a feature sets based on parameters. These lagged features you are creating here are more for models like ARIMA which can accept a regressor but won't have a multivariate lag unless this future regressor is passed.Try passing
summarize=None
tocreate_regressor
. By default it lags all features and then aggregates, down to 10, to keep things faster and cleaner. Removing the summarization will make it more clear what is being done.I don't quite understand what you are asking with
how …