https://github.com/nixtla logo
s

Søren Roesgaard Nielsen

06/05/2023, 6:40 PM
Slightly of topic but maybe someone have a good answer 😊 I am in the process of making a ML model for the next 24 hours each day. What is the best way to make this? A single model with hour as a cyclic feature (a lot easier to maintain since it is only a single model) or make 24 individual models, one for each hour? I am currently using xgboost and lightgbm
k

Kevin Kho

06/05/2023, 7:59 PM
Using the MLForecast library, you can just define the
freq
. And then you can just do:
model.predict(24)
to predict the next 24 hours. It can be one model, and probably should be. See this
s

Søren Roesgaard Nielsen

06/07/2023, 5:30 AM
Thanks I Will look into it 😊
2 Views