Changelog • designbgm

designbgm 0.3.0

New planning method "BSDA" (Bayesian structural design analysis) in design(). It recommends a sample size at which edge selection reaches a target sensitivity or specificity with a given power, using the probit inversion of a simulated power curve.
ggm_parameters() gains a pip argument (prior inclusion probabilities), supplied as a single probability or a matrix. Required by BSDA and carried through elicit_prior().
bsda_control() is an helper control list where the user can specify: the sampler, model-fit, and search settings for BSDA planning.
validate() now supports BSDA plans, reporting the achieved power at the recommended sample size with a confidence interval.
design(method = "BFDA") gains an edge argument to plan around a specific edge directly (1-based indices into K/G), bypassing rho_quantile-based automatic selection.
New generic power_curve(), for method = "DPIR" (Pr(DPIR > threshold) at a grid of sample sizes, both globally and averaged across off-diagonal parameters — unlike design()’s "pw" target, which is the weakest parameter, not the average), method = "BFDA" (power and error rate at a grid of sample sizes, for a single edge), and method = "BSDA" (Pr(sensitivity/specificity >= target) at a grid of sample sizes, over the whole graph), instead of searching for n*. print() and plot() functions are available for all three methods. "DPIR" is the default method, as in design().

print() methods for ggm_parameters and ggm_elicited now display the prior inclusion probabilities (range) when supplied.
Internal C++ routines standardized under the cpp_ naming convention:
- Renamed internal helpers (e.g., power_at_n() to cpp_power_at_n()).
- Added cpp_-prefixed wrapper functions to convert C++ structs to Rcpp::List for R export.

New exported function constrain_precision_to_graph(), which projects a precision matrix onto a fixed undirected graph and returns the precision matrix constrained to the graph’s zero pattern (where entries at non-edges are set to zero).
design.ggm_elicited() now includes an nsim_bf argument controlling the number of Monte Carlo simulations used in the Bayes-factor computation for the sparse (G-Wishart) BFDA path. Defaults to 1000L, matching the underlying C++ default. Finally, the value is recorded in the returned design object’s call_info for sparse BFDA designs.
The DPIR and dense BFDA design routines now report timing information, which in 0.1.0 was available only for the sparse BFDA path. Durations are returned in the design object: duration_bisection_global and duration_bisection_pw for DPIR, duration_h0 and duration_h1 for BFDA.