bsts AR 3 model produces different numbers of samples than other models about episoon HOT 9 CLOSED

This isn't the most informative issue I have
seen....

from episoon.

the issue is that for the AR 3 model some outputs are not produced. And therefore the lengths of the raw_forecast samples differ as a symptom of an underlying problem.

from episoon.

hm okk the bug only happens sometimes. so I assume it has something to do with the AR 3 fit
I don't think it is an issue of the input format. I retried with

obs_rts <- EpiSoon::example_obs_rts %>%
  dplyr::mutate(timeseries = 1, sample = 1) %>%
  rbind(EpiSoon::example_obs_rts %>%
          dplyr::mutate(timeseries = 2, sample = 1))
  
obs_cases <- EpiSoon::example_obs_cases %>%
  dplyr::mutate(timeseries = 1, sample = 1) %>%
  rbind(EpiSoon::example_obs_cases %>%
      dplyr::mutate(timeseries = 2, sample = 1))

and the issue still persists sometimes

from episoon.

Can you make a reprex and will review. Even better enter the debug and step through. :)

from episoon.

at least a bit more minimal and reproducible:

models <- list("AR 3" =
                 function(...) {EpiSoon::bsts_model(model =
                                                      function(ss, y){bsts::AddAr(ss, y = y, lags = 3)}, ...)})

obs_rts <- EpiSoon::example_obs_rts %>%
  dplyr::mutate(timeseries = 1, sample = 1)

for (i in 1:10) {
  tmp <- iterative_rt_forecast(rts = obs_rts, model = models[[1]], 
                               horizon = 7, samples = 10)
  print(nrow(tmp))
}

from episoon.

awesome - ty

from episoon.

So the issue is bsts models randomly fail right - is that not a known issue? or am I confused here?

from episoon.

So the issue is bsts models randomly fail right

I think basically yes.

from episoon.

So I was thinking should put in a proportion successful to one of the summary calls so this can be checked. Will look at doing tomorrow? Thanks for exploring :)

from episoon.