SimInf (version 11.1.0)

SimInf_abc-class: Class SimInf_abc

Description

Storage class for the results of an Approximate Bayesian Computation (ABC) parameter estimation using Sequential Monte Carlo (SMC). The SimInf_abc class holds the model definition, prior distributions, accepted parameter values (particles), weights, distances, and convergence diagnostics.

Slots

model

A SimInf_model object containing the model structure (transitions, compartments, etc.) for which parameters are being estimated.

priors

A data.frame defining the prior distributions for the parameters. It contains four columns:

• parameter: The name of the parameter in the model.

• distribution: The prior distribution type. Valid values are "gamma", "lognormal", "normal", or "uniform".

• p1: The first hyperparameter:

  • "gamma": shape

  • "lognormal": meanlog (mean on the log scale)

  • "normal": mean

  • "uniform": lower bound

• p2: The second hyperparameter:

  • "gamma": rate

  • "lognormal": sdlog (standard deviation on the log scale)

  • "normal": sd (standard deviation)

  • "uniform": upper bound

target

Character vector ("gdata" or "ldata") that determines if the ABC-SMC method estimates parameters in model@gdata (global data) or in model@ldata (local data).

pars

An integer vector with the indices of the parameters in target that are being estimated.

nprop

An integer vector with the number of simulated proposals (particles) generated in each generation.

fn

A function used to calculate summary statistics from the simulated trajectory and compute the distance for each particle. See abc for details on the required function signature.

tolerance

A numeric matrix (number of summary statistics \(\times\) number of generations) where each column contains the tolerances for a generation and each row contains a sequence of gradually decreasing tolerances.

x

A numeric array (number of particles \(\times\) number of parameters \(\times\) number of generations) with the parameter values for the accepted particles in each generation. Each row is one particle.

weight

A numeric matrix (number of particles \(\times\) number of generations) with the weights for the particles x in the corresponding generation.

distance

A numeric array (number of particles \(\times\) number of summary statistics \(\times\) number of generations) with the distance for the particles x in each generation. Each row contains the distance for a particle and each column contains the distance for a summary statistic.

ess

A numeric vector with the effective sample size (ESS) in each generation. The effective sample size is computed as $$\left(\sum_{i=1}^N\!(w_{g}^{(i)})^2\right)^{-1},$$ where \(w_{g}^{(i)}\) is the normalized weight of particle \(i\) in generation \(g\).

init_model

An optional function that, if non-NULL, is applied before running each proposal. The function must accept one argument of type SimInf_model with the current model of the fitting process and return a modified model. This function can be useful to specify the initial state of u0 or v0 of the model before running a trajectory with proposed parameters.

See Also

abc for the main ABC function and continue_abc for continuing an ABC run.