Good morning, Thank you for the support! This is ...
# mlforecast
b
Good morning, Thank you for the support! This is a great community 😊 I would love to integrate Diebold-Mariano and Giacomini-White tests (https://epftoolbox.readthedocs.io/en/latest/modules/statest/gw.html) into Nixtla to evaluate model performance. Additionally, I'd find hourly comparison plots like the attached image very useful. I have a quick question: Does using
max_horizon
with
lag 1
introduce data leakage? (see second image) Grazie mille!
Additional question: is the calibration window taken from train part or from the test part?
j
It shouldn't leak, the future targets are added to each row. The calibration in CV is taken from the train part (as it's done in fit)
1
b
Grazie!