How to train based on multiple series?

[torronen]

I have measurements from a few thousands of household for 1-2 years (varying length per household). Does it make sense to train based on all these series, does Neural Prophet benefit from it, and how to I give multiple series for netural prophet?

Background
Goal: make electricity consumption forecast for a single household

I would like to tackle 2 issues:

1. single household may have too much noise (someone forgot fridge door open). I could predict the average, but :
2. different household may have different kinds of sesonality. For example a houshold with a stay-at-home mom may have very different charastericis from someone who visits his parents every weekend.

My model should detect the seasonality type (2) based on historic data, and then make the best possible prediction, maybe utilizing the noise (1) in upper and lower thresholds. I

[torronen]

I believe the feature I am looking for is called "global model" vs "local model" vs panel data. Are there any samples on how to feed the data?

[torronen]

I'll next try

1. create dataframe for each series as normel
2. combine them with join_dataframes ( d07b795 )

any tips appreciated