extensions - Allow for prediction of unobserved years

[sebdalgarno]

The contractor will extend the bboutools survival, recruitment, and population growth models to allow for prediction of unobserved years.

[sebdalgarno]

do we require that the user uploads data with years and NA values, or do we provide an option to specify years for which estimate is required and if no data, then predict

[sebdalgarno]

im a bit unclear on how the landscitech implementation currently works.
for example, if I do

  surv <- bboudata::bbousurv_a
  surv <- surv[!(surv$Year %in% 1998:2002),]
  surv <- rbind(surv, data.frame(PopulationName = "A", Year = 1998, Month = 1, 
                                 StartTotal = NA_integer_, MortalitiesCertain = NA_integer_, 
                                 MortalitiesUncertain = NA_integer_))
  fit <- bb_fit_survival(surv, allow_missing = TRUE)

I get

[sebdalgarno]

I think probably a more elegant solution is to implement prediction of unobserved years in the bb_predict_x functions i.e. with new_data, rather than requiring the user to add rows of NA data to dataset

[sebdalgarno]

from Josie:

I appreciate that adding the predictions for unobserved years at the predict stage will improve speed and flexibility.

However, I would also like to retain some possibilities that the current implementation allows:

1. the option to "fit" models with no data to examine the prior predictions
2. the option to fit random effects models given less than 5 years of observations
3. the option to examine and use mcmc samples from an unobserved year