bayes_R2: compute the Bayesian R square for a greta regression model

In lionel68/greta.checks: Perform Checks on greta Models and Arrays

compute the Bayesian R square for a greta regression model

Description

Compute a Bayesian version of R-square for regression models (GLMs ...)

Usage

## S3 method for class 'mcmc.list'
 bayes_R2(y, pred, draws, summary = TRUE, probs = c(0.1, 0.9))

Arguments

y	a greta array, the response variables
pred	a greta array, the linear predictor
draws	a greta_mcmc_list, posterior draws as returned from calling greta sampling algorithm (ie mcmc)
summary	a logical, if TRUE (default) the function output summary statistics (mean, sd, 80% credible intervals) for the R2, if FALSE the raw values are returned
probs	a vector of two numeric specifying the lower and upper limits for the credible intervals (default to 0.1, 0.9), only used if summary=TRUE

Details

See https://github.com/jgabry/bayes_R2/blob/master/bayes_R2.pdf for a description of the computation. Note that R2 only univariate models are supported. This function is largely inspired from brms::bayes_R2.

Value

If summary=TRUE a 1 x C matrix is returned (C = length(probs) + 2) containing summary statistics of Bayesian R-squared values. If summary = FALSE the posterior samples of the R-squared values are returned as a numeric vector of length S (S is the number of samples)

Examples

## Not run: 
intercept <- normal(0, 1)
slope <- normal(0, 1)
sd_resid <- cauchy(0, 1, truncation = c(0, 100))

x <- runif(100)
y <- as_data(rnorm(100, 1 + 2 * x, 1))

pred <- intercept + slope * x
distribution(y) <- normal(pred, sd_resid)

m <- model(intercept, slope, sd_resid)
drr <- mcmc(m)

bayes_R2(y, pred, drr)

## End(Not run)

lionel68/greta.checks documentation built on July 30, 2022, 11:39 a.m.