https://github.com/nixtla logo
a

Andrew Doherty

07/26/2023, 10:45 AM
Hi @Max (Nixtla), in your response you mention local scalers for MLForecast but I can't find this in the documentation/code. I must be missing it. Could you share where this is? Thanks.
m

Max (Nixtla)

07/26/2023, 2:07 PM
Hi Andrew! In mlforecast you have to specifically tell the software which scaler to use. You can do so through this argument: target_transforms=[LocalStandardScaler()] See line 14 on this notebook: https://www.kaggle.com/code/fedegr/mlforecast-functime/notebook
@José Morales: maybe we should write a little guide or something.
j

José Morales

07/26/2023, 5:43 PM
Yeah I have a guide for target transforms almost done, will add it soon
a

Andrew Doherty

07/27/2023, 8:23 AM
Excellent, thanks a lot @Max (Nixtla) and @José Morales. Do you plan on implementing for exogenous features? To enable sliding/expanding window standard scaling for example.
j

José Morales

07/27/2023, 9:04 PM
This is already possible by doing something like here https://github.com/Nixtla/mlforecast/issues/124#issue-1683960264 But we will add a built in way to do it soon
a

Andrew Doherty

07/31/2023, 9:41 AM
OK nice one. Thanks again.