Approximate Bayesian Computation for Stochastic Model

Runs an Approximate Bayesian Computation (ABC) procedure using a stochastic model and a set of prior parameter draws.

Usage

abc_stochastic_model(
  sm,
  priors,
  target,
  stat_func,
  states = NULL,
  tol = 0.1,
  method = "rejection",
  seed = 123
)

# S3 method for class 'abc_stochastic_model'
plot(x, ...)

Arguments

sm: An object of class stochastic_model.
priors: A data frame or tibble() of prior parameter values. Each row should correspond to one simulation, and columns should match the parameter names expected by update_parameters().
target: A named vector or data frame of observed summary statistics to match against simulated summary statistics.
stat_func: A function that computes summary statistics from a simulation result. Should accept a model object and return a named numeric vector or tibble() row.
states: Optional. A data frame or tibble() of initial states. Each row corresponds to a complete model state e.g. for an SIR model each row has three columns
tol: A numeric tolerance (between 0 and 1) for the ABC rejection algorithm. Defaults to 0.1.
method: A character string indicating the ABC method to use. Defaults to "rejection". Passed directly to abc::abc().
seed: An integer specifying the random seed for reproducibility. Defaults to 123.
x: An object of class abc_stochastic_model.
...: list of additional plot arguments. param - if type is "prior_posterior" then parameter to plot state - if type is "simulations" then state to plot

Value

An object of class "abc_stochastic_model" containing:

model: The original stochastic model.
fit: An object returned by abc::abc() representing the ABC posterior.
priors: The input prior values used for simulations.
posteriors: The output posterior values used for simulations.
prior_predictive: The summary statistics defined in stat_func drawn from the prior distribution.
state: Final state of model.

A ggplot object

Details

This function performs ABC by:

Sampling parameter combinations from priors
Updating the stochastic_model with each sampled parameter set
Running the model and computing summary statistics via stat_func
Comparing simulated statistics to target using the abc::abc() function

Parallelization is handled using furrr::future_map(), allowing reproducible simulation across parameter sets using a fixed seed.

Methods (by generic)

plot(abc_stochastic_model): Plot of either prior and posterior or projections

Examples

reactions <- list(
  infection = list(
  transition = c("I" = +1),
  rate = function(x,p,t){ p$beta })
)
example_scenario <- list(
  params = list(beta = 0.1),
  initial_states = list(I = 0),
  sim_args = list(T = 10)
)
sm <- stochastic_model(reactions,example_scenario)
priors <- expand.grid(beta = seq(0.1, 0.5, length.out = 10))
target <- c(total_infected = 50)
abc_result <- abc_stochastic_model(
  sm,
  priors = priors,
  target = target,
  stat_func = function(sim) data.frame(total_infected = sim[nrow(sim), "I"]),
  tol = 0.2
)