winedarksea / AutoTS

Automated Time Series Forecasting
MIT License
1.1k stars 100 forks source link

NaN in prediction #87

Closed treexterr closed 2 years ago

treexterr commented 3 years ago

Hi. Thx for great job!

I've got NaN in prediction, while the prediction horizon started one timestamp earlier.

What I did wrong?

http://joxi.ru/82Q1PXDcVR69B2

winedarksea commented 3 years ago

I'm very reluctant to click on that link (I've never seen joxi.ru before), sorry... Maybe there is some other way to provide information?

Best guess is that it is a problem with how your data is being loaded. Check model.df_wide_numeric where model is your fit AutoTS object. Does that look correct to you?

If it is an error with modeling, there's usually a simple way to check: simply run AutoTS again. Bugs in models are unlikely to be seen consistently, as different models are usually chosen on a new run.