Nixtla's Open Source Time Series Ecosystem.

Nixtla Community

I see the documentation for transfer learning using MLforecast. Does this also work if I use additional features (no date features)?

Yes, it should also work. You have to provide both the new_df and X_df. Are you getting an error?

new_df is like the "train" dataframe and X_df the future

crazy :slightly_smiling_face: No error. i didnt try because i thought it would not work, but wanted to ask before i would try stuff that would never work. but pretty cool :slightly_smiling_face:

<@U03UX9KR75W> to come back to this: so i have to know until when the model was trained and only use ds until that day for new_df? what if i updated the model? i guess i can just adapt the dates due to the updating, or is updating in combination with new_df not possible?

Yes, at the moment you have to make sure you're using future dates in the update method. These are available in the `mlf.ts.last_dates` , which is a series with ids and last times, so you can join your dataframe with that and filter where your dates are greater than those.