In greta-dev/greta: Simple and Scalable Statistical Modelling in R

Exponential-normal prior

In this example we infer the parameters of one-variable Bayesian linear regression model using an exponential-normal prior. A compound exponential-normal prior can be interpreted like an equivalent to the frequentist LASSO. The exponential-normal prior yields a posterior that is pooled towards zero. An exponential-normal prior, or equivalently a Laplace prior, is consequently often chosen when a sparse solution is assumed, which, for instance, is a natural scenario in many biological settings.

# variables & priors
int <- variable()
sd <- inverse_gamma(1, 1)

lambda <- gamma(1, 1)
tau <- exponential(0.5 * lambda**2)
coef <- normal(0, tau)

# linear predictor
mu <- int + coef * attitude$complaints

# observation model
distribution(attitude$rating) <- normal(mu, sd)