FlywayNet (version 0.1.0)

reestimate_migration_from_MCEMruns: Estimates migration parameters from multiple MCEM runs

Description

Estimates migration parameters using multiple runs of MCEM. Since MCEM is a local optimization process, different initialisation of the parameters can lead to different estimates. Based on log likelihood, this function searches a posteriori for the convergence of each MCEM run and returns the results of the best MCEM run. Convergence of a MCEM run is considered reached when difference in log likelihood is below loglikelihood_threshold during nb_iterations_threshold consecutive iterations.

Usage

reestimate_migration_from_MCEMruns(
  mcem_runs,
  loglikelihood_threshold = 10,
  nb_iterations_threshold = 5
)

Arguments

mcem_runs

List of mcem runs as returned by estimate_migration_MCEM

loglikelihood_threshold

Threshold on loglikelihood difference between two consecutive iterations

nb_iterations_threshold

Number of consecutive iterations with a loww difference in loglikelihood.

Value

The migration structure given with a new attribute

estimation_method which is a structure with 3 attributes:

. name = "MCEM_combined",

. settings, a structure with attributes: loglikelihood_threshold, nb_iterations_threshold.

. output, a structure with attributes:

- transition_law_param: estimated transition parameters,

- sojourn_law_param: estimated sojourn parameters,

- index_run_max: maximum number of iteration for convergence,

- nb_simulations_total: total number of simulations,

- nb_simulations_at_convergence: number of simulations useful at convergence,

- summary: a summary.

Examples

# NOT RUN {
migr1 <- generate_toy_migration()
estimated_migr1 <- estimate_migration_MCEM(migr1, MC_algo="MH",   
                                          log_transitions = TRUE,
                                          log_sojourns = TRUE,
                                          log_loglikelihood = TRUE) 
migr2 <- generate_modified_migration(migr1, 
  sojourn_mode=c(rep("unif", 4), "no"),
  sojourn_domain=c(1,5),
  transition_mode="unif")
estimated_migr2 <- estimate_migration_MCEM(migr2, MC_algo="MH",
                                          log_transitions = TRUE,
                                          log_sojourns = TRUE,
                                          log_loglikelihood = TRUE) 
estimated_migr = reestimate_migration_from_MCEMruns(list(estimated_migr1, estimated_migr2))
print( estimated_migr$estimation_method$output$transition_law_param )
# }