cmrdesign 0.1.0
First CRAN-oriented release of cmrdesign.
- Added applied Conditional Minimax Regret rules for two-arm,
unbounded two-arm, shared-control multi-arm, stratified,
multiple-outcome, and proxy outcome designs.
- Added pilot/main-wave sample-size planning tools from Appendix E of
the accompanying paper.
- Added R/Python cross-language fixtures, reference validation,
vignettes, and compact print/summary methods for core CMR results.
- Added
realize_allocation() to turn CMR target shares
into integer main-wave counts and recompute the rounded-design
certificate when possible.
- Added allocation cross-language fixtures and vignette/examples
coverage for integer count realization.
- Hardened
realize_allocation() input validation for
named length-one targets, fixed stratum counts, zero-count realized
arms, and certificate recomputation diagnostics.
- Maurer–Pontil bounded-outcome fits now report the raw Bessel sample
variance in
vhat and keep projected unit-interval variances
in the method-specific statistic$projected_vhat field.
- Added explicit warnings when
normalize = TRUE uses
pilot min/max support bounds, clarifying that this convenience path is
exploratory rather than finite-sample guarantee-bearing.