prediction dependent on the order of the added sesonality

luke14free / pm-prophet

GAM timeseries modeling with auto-changepoint detection. Inspired by Facebook Prophet and implemented in PyMC3

327 stars 42 forks source link

prediction dependent on the order of the added sesonality #23

Open BjorHart opened 4 years ago

BjorHart commented 4 years ago

hi, I have discovered that altering the order of which I add the weekly and monthly seasonalities to my model affects the outcome. The two figures below show weekly added before monthly, and monthly added before weekly. What causes this order depencency?

7then30 30then7

t8ch commented 4 years ago

Hi. Could you provide give some details about the data and inference parameters. By any chance, have you checked whether the differences might be due to convergence issues (stochastic in nature)?

luke14free commented 4 years ago

I am curious to see if this persists after my latest fixes in seasonality, can you try re-running this @BjorHart ? Best

BjorHart commented 4 years ago

hi! I will give it a go and get back to you @luke14free .