I would like to do hyperparameter optimisation while using the awesome fable package(here fable.prophet) from tidyverts
Using the example of fable.prophet like as mentioned at https://github.com/mitchelloharawild/fable.prophet.
fit <- cafe %>%
model(
prophet = prophet(Turnover ~ season("year", 4, type = "multiplicative"))
)
Here the parameters have been already decided in the example. But I would like to arrive at the parameters myself after trying a range of values.
Like say
type = c('additive', 'multiplicative') ,
changepoint_prior_scale = c(0.005, 0.05, 0.5, 5) ,
changepoint_range = c(0.8, 0.9),
growth = c('linear', 'logistic')
Since these parameters may not be the same for my dataset. I would like to do hyperparameter tuning/optimisation using tune() from tidymodels or use some other method to efficiently go through a range of parameter values before arriving at the optimal ones.
The closest I could find was a discussion here https://community.rstudio.com/t/hyperparameter-tuning-facebook-prophet-in-r/97395 or modeltime but I am looking for hierarchical/grouped/nested timeseries forecast with reconciliation hence I prefer to use only fable.prophet.
I can solve this manually by going through each group in the data seperately but this has to be redone every 6-12 months for something like 30 groups at least if not 10 more subgroups of data. (Country, Platform,Product, etc)
It would be great if anyone could point to any already existing way or a clever way to solve this.
Thanks