README.md

In finnlindgren/multiprobit: Bayesian Multivariate Probit Regression

multiprobit

The goal of multiprobit is to perform fast Bayesian inference for multivariate probit models. The method uses a latent Gaussian variable parameterisation of the correlation matrix, and numerical optimisation and integration to find the posterior distributions of the model coefficients and correlation matrix.

Installation

To install the latest stable release version of multiprobit from github, use

remotes::install_github("finnlindgren/multiprobit", ref = "stable")

To install the development version of multiprobit from github, use

remotes::install_github("finnlindgren/multiprobit", ref = "devel")

Example

This is a basic example which shows you how to solve a common problem:

if (interactive()) {
  library(multiprobit)

  N <- 6
  d <- 2
  J <- 2

  set.seed(1L)
  X <- cbind(1, matrix(rnorm(N * (J - 1)), N, J - 1))
  B <- matrix(0.5, J, d)
  Y <- matrix(rnorm(N * d, mean = as.vector(X %*% B)) > 0, N, d)
  df <- d + 1
  prec_beta <- 0.1

  model <- mp_model(
    response = Y, X = X,
    df = df, prec_beta = prec_beta
  )
  opt <- multiprobit(
    model = model,
    options =
      mp_options(
        gaussint = list(max.threads = 1),
        strategy = "stepwise"
      )
  )
}

finnlindgren/multiprobit documentation built on June 20, 2020, 6:12 a.m.