To use waspr
the user first needs to load the package as
follows:
The user must provide a 3-dimensional array with posterior samples
for all parameters for each subset posterior (rows = subset posteriors,
columns = parameters, slices = samples). The amount of parameters and
samples must be equal for each subset posterior, these posterior samples
may be obtained from any type of MCMC algorithm. waspr
provides an example array with posterior samples for 8 parameters for 8
subset posteriors, pois_logistic
, that will be used to
illustrate the functionality of the package.
The main function wasp()
runs the swapping algorithm,
combines its output and computes the Wasserstein barycenter. It has four
arguments, mcmc
, that specifies the 3-dimensional array
with samples for each subset posterior, and optional arguments
par.names
, that can be used to specify parameter names,
iter
to specify the maximum number of iterations of the
swapping algorithm, acc
to specify the accuracy of the
swapping algorithm and out
to indicate whether the results
per iteration of the swap algorithm should be printed.
out <- wasp(pois_logistic,
iter = 10,
acc = 0.001,
par.names = c("beta_s", "alpha_l", "beta_l",
"baseline_sigma", "baseline_mu",
"correlation", "sigma_s", "sigma_l"))
wasp()
prints the iteration number and cost function
value of the swapping algorithm. The out
object is of class
wasp
and contains several elements. To obtain the
Wasserstein barycenter of the subset posteriors a user can specify
out$barycenter
. This returns a matrix of posterior samples
(rows) for all parameters (columns) of the full data posterior. A
summary of the approximation of the full data posterior is available
through summary(out)
.
summary(out)
#> mean mode sd LB HPD UB HPD
#> beta_s 0.5527601 0.5518034 0.10988949 0.36598187 0.7896041
#> alpha_l 2.6811079 2.6959176 0.19199304 2.30380675 3.0295802
#> beta_l 0.7508520 0.7339988 0.21631011 0.37281283 1.1740767
#> baseline_sigma 0.3563222 0.3811609 0.06859910 0.21910807 0.4870079
#> baseline_mu -0.8008872 -0.7516167 0.10867533 -1.01168299 -0.5944583
#> correlation 0.1732170 0.1392670 0.07437737 0.02824474 0.3059979
#> sigma_s 1.7225455 1.7535499 0.17920847 1.40126462 2.0610585
#> sigma_l 1.2190297 1.2612822 0.07558163 1.06768047 1.3569757