rbiips-package: Bayesian inference with interacting particle systems

In biips/rbiips: Bayesian Inference with Interacting Particle Systems

Description

rbiips is an interface with the Biips C++ libraries for analysing Bayesian graphical models using advanced particle methods.

Details

Biips is a general software for Bayesian inference with interacting particle systems, a.k.a. sequential Monte Carlo (SMC) methods. It aims at popularizing the use of these methods to non-statistician researchers and students, thanks to its automated “black box” inference engine. It borrows from the BUGS/JAGS software, widely used in Bayesian statistics, the statistical modeling with graphical models and the language associated with their descriptions.

See the Biips website for more information.

The typical workflow is the following:

• Define the model in BUGS language (see the JAGS User Manual for help) and the data.

• Add custom functions or distributions with biips_add_function and biips_add_distribution.

• Compile the model with biips_model

• Run inference algorithms:

  • Analyse sensitivity to parameters with biips_smc_sensitivity.

  • Run SMC filtering and smoothing algorithms with biips_smc_samples.

  • Run particle MCMC algorithms with biips_pimh_samples or biips_pmmh_samples.

• Diagnose and analyze the output obtained as smcarray and mcmcarray objects with biips_diagnosis, biips_summary, biips_density, biips_hist and biips_table

Author(s)

• Adrien Todeschini

• François Caron

• Marc Fuentes

rbiips development was initiated by the research team ALEA at Inria Bordeaux Sud-Ouest.

References

Adrien Todeschini, François Caron, Marc Fuentes, Pierrick Legrand, Pierre Del Moral (2014). Biips: Software for Bayesian Inference with Interacting Particle Systems. arXiv preprint arXiv:1412.3779. URL http://arxiv.org/abs/1412.3779

See Also

biips_add_function, biips_add_distribution, biips_model, biips_smc_sensitivity, biips_smc_samples, biips_pimh_samples, biips_pmmh_samples, smcarray, mcmcarray, biips_diagnosis, biips_summary, biips_density, biips_hist, biips_table, Biips website, JAGS User Manual

Examples

#' # Add custom functions and distributions to BUGS language
#' Add custom function `f`
f_dim <- function(x_dim, t_dim) {
  # Check dimensions of the input and return dimension of the output of function f
  stopifnot(prod(x_dim) == 1, prod(t_dim) == 1)
  x_dim
}
f_eval <- function(x, t) {
  # Evaluate function f
  0.5 * x + 25 * x/(1 + x^2) + 8 * cos(1.2 * t)
}
biips_add_function('f', 2, f_dim, f_eval)

#' Add custom sampling distribution `dMN`
dMN_dim <- function(mu_dim, Sig_dim) {
  # Check dimensions of the input and return dimension of the output of
  # distribution dMN
  stopifnot(prod(mu_dim) == mu_dim[1], length(Sig_dim) == 2, mu_dim[1] == Sig_dim)
  mu_dim
}
dMN_sample <- function(mu, Sig) {
  # Draw a sample of distribution dMN
  mu + t(chol(Sig)) %*% rnorm(length(mu))
}
biips_add_distribution('dMN', 2, dMN_dim, dMN_sample)

#' # Compile model
modelfile <- system.file('extdata', 'hmm_f.bug', package = 'rbiips')
stopifnot(nchar(modelfile) > 0)
cat(readLines(modelfile), sep = '\n')

data <- list(tmax = 10, p = c(.5, .5), logtau_true = log(1), logtau = log(1))
model <- biips_model(modelfile, data, sample_data = TRUE)

#' # SMC algorithm
n_part <- 100
out_smc <- biips_smc_samples(model, c('x', 'c[2:10]'), n_part, type = 'fs',
                             rs_thres = 0.5, rs_type = 'stratified')

biips_diagnosis(out_smc)
biips_summary(out_smc)
par(mfrow = c(2, 2))
plot(biips_density(out_smc$x, bw = 'nrd0', adjust = 1, n = 100))
plot(biips_table(out_smc[['c[2:10]']]))

biips/rbiips documentation built on Nov. 28, 2020, 2:12 p.m.