Fit nimue to a given set of death data via the particle and optimisation exploration approach.

For a given number of samples from given parameter uncertainty or distribution function, this approach iteratively fits the Rt trend to the provided death data and returns a nimue_simulation object for future usage in scenario modelling.

Usage

rt_optimise(
  data,
  distribution,
  squire_model,
  parameters,
  start_date,
  parallel = FALSE,
  rt_spacing = 14,
  k = 2,
  n_particles = 14,
  initial_infections_interval = c(5, 500),
  rt_interval = c(0.5, 20),
  dt = NULL
)

Arguments

data: A data frame of deaths occuring over a given time frame. Given in the format: deaths(integer), date_start(date), date_end(date). Must have at least one death period in each set of Rt trend changes (i.e. a 14 day period by default) and no period can overlap these changes.
distribution: A list of samples(a list) with names specifying parameters.
squire_model: A model object of the desired type to use, i.e. squire, nimue.
parameters: A list of parameters to keep the same across all samples.
start_date: Date when the epidemic begins, parameters and distribution should be formatted relative to this date, where necessary.
parallel: Run each sample concurrently, uses the future::plan set by the user. Default = FALSE.
rt_spacing: Number of days between each Rt trend, default = 14 days.
k: Control the dispersion on the negative binomial likelihood, default = 2.
n_particles: How many particles to explore uniformly the interval of initial infections, default = 7.
initial_infections_interval: The range of initial number of infections to explore, default = c(5, 500).
rt_interval: The range of values that Rt can take, default = c(0.5, 20).
dt: If the passed squire_model has a difference model attached, this is the step size we shall use. The difference model is only used if dt is non-null and squire_model$odin_difference_model is non-null. Defaults to NULL.

Value

An object of type rt_optimised, (model type).

Details

NOTE: For death curves with periods of 0's rt_interval's lower bound must be greater than 0 else it will likely fail to overcome a low infective population and Rt will tend to some unrealistically large number.

This function is progressr enabled, so progressr::handlers(global = TRUE) can be used to view progress through the samples.