Skip to contents

Changelog

rxsim 0.3.0

This release is mostly about paying down debt. The timer got a simpler internal representation, fixed designs got a faster execution path, and two long-deprecated helpers finally came out. If you are upgrading from 0.2.0, the removals below are the only changes that should require edits on your side.

Breaking changes

  • add_timepoints() and Timer$get_timepoint() have been removed (#107). Both predate Timer$add_schedule(), which does the same job in one call and accepts the output of stochastic_schedule() and deterministic_schedule() without reshaping.

    # before
add_timepoints(timer, plan)

# after
timer$add_schedule(plan)

    If you built timepoints by hand, pass a data frame with time, arm, enroll, and drop columns straight to add_schedule(). Note that enroll and drop are subject counts and must be integers (3L, not 3).

  • Timer$timelist is now a plain data frame rather than a list of lists (#107). Code that reassembled it with do.call(rbind, timer$timelist) can read the field directly. Anything that indexed into it positionally will need updating.

  • stochastic_schedule() and deterministic_schedule() are now genuine drop-in replacements for each other. Both return a plain data.frame sorted by arm and then time; previously deterministic_schedule() returned a tibble with a different row order, which meant swapping one for the other could quietly change downstream behaviour.

New features

  • Fixed designs (adaptive = FALSE, the default on Trial$new() and replicate_trial()) now run through a dedicated fast path that precomputes enrollment and dropout times and takes cheap prefix slices for each snapshot (#108). A companion optimisation skips leading timepoints where no condition can possibly fire (#109). Results are identical to the adaptive loop; only the runtime differs.

Bug fixes

  • deterministic_schedule() no longer fails when a sample-size correction lands before the first dropout period boundary. findInterval() returns zero in that case, and the resulting empty vector crashed the schedule assembly. Corrections are now attributed to the first period, which is what the right-closed period semantics imply.

  • Trigger expressions combined with | now handle NA guards correctly, and quosure generation for composed triggers was tightened up (#106).

  • n_readouts > 1 is now handled consistently across both execution paths.

Documentation

  • A thorough accuracy pass over the README and all five vignettes. Several column tables, code examples, and cross-references had drifted behind earlier API changes; these now match the shipped behaviour. The README quick-start example and the assembly walkthrough in Core Concepts are executed at build time, so they cannot silently rot again.

  • The trigger reference is complete: notna_trigger(), col_trigger(), timed_count_trigger(), and trigger_by_events() are now documented alongside the primitives they complement, and the fire-once default (max_triggers = 1L) is stated explicitly with a runnable example.

rxsim 0.2.0.9000

New features

  • New event-driven trigger, trigger_by_events(), fires an analysis once a target number of time-to-event endpoints has been observed — the natural way to schedule looks in oncology and other event-driven settings (#99). It is built from three new lower-level primitives that are also exported for custom rules: notna_trigger(), col_trigger(), and timed_count_trigger().

Improvements

  • Added statistical ground-truth tests covering the stochastic assignment paths, plus verification that the adaptive and fixed execution paths produce identical results.
  • Unit test coverage extended to the helper functions.

rxsim 0.1.7.9000

Breaking changes

  • The five worked examples have moved to a separate rxsim-gallery site (#103). They were the only reason rxsim suggested RBesT, DoseFinding, BayesianMCPMod, multcomp, survival, MASS, and several others; dropping them removes a substantial installation burden from a package whose core has always been small. The examples themselves are unchanged and remain online.

rxsim 0.1.6.9000

A large API consolidation. This is the release where the trigger system took its current shape.

Breaking changes

New features

  • A new Condition class separates trigger logic from the Timer, which is now purely the trial clock. Conditions are passed to Trial$new(conditions = ...).
  • Composable trigger objects: value_trigger(), count_trigger(), enroll_trigger(), and calendar_trigger() can be combined with & and | to express rules such as “half enrolled and past month six” without writing filter expressions by hand.
  • Population now validates that all arms declare the same number of readouts, catching a class of mismatch that previously surfaced as a confusing error deeper in the run loop.

Improvements

  • collect_results() rewritten for clarity and speed.
  • Test coverage added for the Trial class.
  • Vignettes reorganised around concepts rather than individual classes.

rxsim 0.1.2.9000

New features

  • Conditions can now be capped so they do not fire repeatedly: cooldown sets a minimum time gap between triggers, and max_triggers sets a hard ceiling (#58). Useful for data-driven checks evaluated at every timepoint.

Improvements

  • A helper for building population data frames, and a polish pass across all worked examples and the getting-started guide.

rxsim 0.1.0.9000

  • First tagged release with the package website published via pkgdown, along with continuous integration running R CMD check on every push.

rxsim 0.0.6.9000

  • Initial version. Establishes the three building blocks the package is still organised around — Population, Timer, and Trial — together with replicate_trial() and run_trials() for generating and executing many independent replicates.