Skip to contents

The RBesT tools are designed to support in the derivation of parametric informative priors, asses design characeristics and perform analyses. Supported endpoints include normal, binary and Poisson.

Details

For introductory material, please refer to the vignettes which include

  • Introduction (binary)

  • Introduction (normal)

  • Customizing RBesT Plots

  • Robust MAP, advanced usage

The main function of the package is gMAP(). See it's help page for a detailed description of the statistical model.

Global Options

OptionDefaultDescription
RBesT.MC.warmup2000MCMC warmup iterations
RBesT.MC.iter6000total MCMC iterations
RBesT.MC.chains4MCMC chains
RBesT.MC.thin4MCMC thinning
RBesT.MC.save_warmupFALSEMCMC warmup samples saving
RBesT.MC.controllist(adapt_delta=0.99,sets control argument for Stan call
stepsize=0.01,
max_treedepth=20)
RBesT.MC.ncp1parametrization: 0=CP, 1=NCP, 2=Automatic
RBesT.MC.init1range of initial uniform \([-1,1]\) is the default
RBesT.MC.rescaleTRUEAutomatic rescaling of raw parameters
RBesT.verboseFALSErequests outputs to be more verbose
RBesT.integrate_argslist(lower=-Inf,arguments passed to integrate for
upper=Inf,adaptive integration of densities (used when
rel.tol=.Machine$double.eps^0.25,RBesT.integrate_method is "adaptive"
abs.tol=.Machine$double.eps^0.25,or for non-normMix densities)
subdivisions=1E3)
RBesT.integrate_prob_eps1E-6probability mass left out from tails if integration needs to be restricted in range
RBesT.integrate_method"GQ"integration method for mixture densities: "GQ" (Gaussian quadrature, deterministic) or "adaptive" (adaptive Gauss-Kronrod). GQ uses Gauss-Hermite for normMix, Gauss-Jacobi for betaMix, and Gauss-Laguerre for gammaMix.
RBesT.GQ_nodes20starting number of Gaussian quadrature nodes (only used when RBesT.integrate_method is "GQ")
RBesT.GQ_rel_tol1E-4relative tolerance target for GQ refinement; the node count is increased until successive estimates agree within max(RBesT.GQ_abs_tol, RBesT.GQ_rel_tol * |I|). Set to a non-finite or non-positive value (e.g. Inf) to disable refinement (single evaluation at RBesT.GQ_nodes).
RBesT.GQ_abs_tol1E-6absolute tolerance floor for GQ refinement
RBesT.GQ_max_nodes240upper cap on the GQ node count during refinement
RBesT.GQ_node_growth2multiplicative growth factor for the GQ node count between refinement steps
RBesT.GQ_on_nonconvergence"adaptive"behaviour when GQ refinement reaches RBesT.GQ_max_nodes without meeting tolerance: "adaptive" (fall through to adaptive integration), "warn" (warn and return best estimate), "error", or "silent"

Version History

See NEWS.md file.

References

Stan Development Team (2020). RStan: the R interface to Stan. R package version 2.19.3. https://mc-stan.org

Author

Maintainer: Sebastian Weber sebastian.weber@novartis.com

Authors:

Other contributors: