https://github.com/nixtla logo
m

Mário Amorim Lopes

04/10/2022, 11:26 AM
Hey everyone! First off, thanks for this very helpful package. Quick question: any way to easily get the fitted model out of croston_classic, cronston_optimized or tsb?
m

Max (Nixtla)

04/10/2022, 4:40 PM
@fede (nixtla) (they/them) can help with that :)
m

Mário Amorim Lopes

04/11/2022, 2:25 PM
Thanks, Max! Hoping to hear from @fede (nixtla) (they/them)!
f

fede (nixtla) (they/them)

04/11/2022, 6:20 PM
Hi @Mário Amorim Lopes! For the moment, these models don’t store additional information, the functions simply take a time series and return the forecasts (this design is highly efficient in parallel tasks). We are currently working on fit-predict classes. 🙂
m

Mário Amorim Lopes

04/11/2022, 6:33 PM
Thanks for the feedback, @fede (nixtla) (they/them) (same name as by second son, btw!). That’s what I suspected by looking at the source code 🙂 Getting the model fit is useful for calculating in-sample error metrics. I’m forecasting intermittent/lumpy products (5k SKUs times ~40 local pharmacies) and I need a way to compare croston, tsb, etc. accuracy.