Skip to contents

Draws parameters from uniform priors, simulates one data set per draw and summarises it, as the first step of inference by approximate Bayesian computation (ABC). What setup returns is drawn once and shared by every simulation, so it should hold what is fixed by the data, such as the observed tree and landscape. A parameter that spans orders of magnitude, such as dispersal, is better drawn on the log scale and transformed back inside simulate.

Usage

abc_table(
  prior,
  simulate,
  n,
  setup = NULL,
  summarise = function(sim, shared) summary_stats(sim, shared$tree),
  seed = 1
)

Arguments

prior

Named list with one numeric vector of length 2 per parameter, the lower and upper limits of its uniform prior.

simulate

Function of scenario, a list with one value per parameter, and shared, what setup returned, which gives a simulation. As in run_scenarios().

n

Number of simulations.

setup

Optional function without arguments that gives what all simulations share.

summarise

Function of sim and shared that gives a named numeric vector. The default is summary_stats(), with shared$tree if there is one.

seed

Integer that sets all the seeds.

Value

An mc_abc_table: a data frame with one row per simulation, the parameters and the statistics, with the attributes prior, parameters and statistics.

See also

abc_fit() to estimate the parameters, abc_recovery() to check whether they can be recovered.

Other inference: abc_choose(), abc_cv(), abc_diagnose(), abc_fit(), abc_ppc(), abc_recovery(), fold_summarise(), observe_detection(), spatial_folds()

Examples

shared_pool <- function() {
  list(tree = treeshapesim::sim_tree_shape(10), landscape = landscape_implicit(8))
}
one_run <- function(scenario, shared) {
  simulate_metacommunity(shared$landscape, species_traits(shared$tree, scenario$lambda),
                         competition_matrix(shared$tree, "none"),
                         dispersal = 10^scenario$log_dispersal, n_steps = 30,
                         seed_steps = 1:3)
}
prior <- list(log_dispersal = c(-3, -0.3), lambda = c(0, 1))
table <- abc_table(prior, one_run, n = 20, setup = shared_pool)
head(table[, 1:5])
#>   log_dispersal    lambda alpha gamma     beta
#> 1    -2.2831266 0.9347052 2.750    10 3.636364
#> 2    -1.9952655 0.2121425 4.500     8 1.777778
#> 3    -1.4532959 0.6516738 6.250     8 1.280000
#> 4    -0.5478390 0.1255551 7.250     8 1.103448
#> 5    -2.4554588 0.2672207 3.625     8 2.206897
#> 6    -0.5743479 0.3861141 4.875     5 1.025641