The package now includes a dedicated routine, bnplasso.spm(), which implements the sparse normal-means problem. The routine returns an object of S3 class, "spmBayes", supported by various class-specific methods.
Other internal functionality improvements.
Latest version available at GitHub.
bnplasso 0.2.1
The package now includes the functions psis.loo() and widely.aic(), which compute the Pareto-smoothed importance sampling leave-one-out information criterion (PSIS-LOO) and the Watanabe–Akaike information criterion (WAIC), respectively.
The functions blasso.lm() and balasso.lm() have been merged into the function, bnplasso.lm(). A new argument, prior, has been introduced in bnplasso.lm(), which specifies the type of shrinkage prior that should be employed. The options are: (1) a nonparametric Bayesian Lasso prior (default), (2) a Bayesian Lasso prior, or (3) a Bayesian adaptive Lasso prior.
The bnplasso.lm() function now supports single-precision floating-point calculations for certain internal routines via the float argument. By default, the function still uses double point precision.
By default, the bnplasso.lm() function now includes an intercept term in the regression function.
The bnplasso.lm() function now returns the log-likelihood of each observation at each MCMC iteration.
If some user-supplied hyperparameters are not provided, the bnplasso.lm() function will now attempt to automatically determine appropriate values for those hyperparameters.
The package now includes the function get.partition(), which recovers the partition of the regression coefficients induced by the nonparametric Bayesian Lasso.
The package now includes the functions coclust.probs() and coclust.point(), which compute and visualize the matrices of co-clustering probabilities and co-clustering point estimates, respectively.
Other internal functionality improvements, including a better handling of numerical instabilities and memory management.