Changelog

Source: NEWS.md

latrend v1.5.12023-03-17

  • Resolved CRAN errors and warnings
  • Alternative implementation of ASW and Dunn index due to removal of clusterCrit package from CRAN (#131, #132)
  • New PAP.adh dataset (#130)
  • Added support to lcmm methods for (parallel) gridsearch (#126)
  • More robust implementation of meta methods, allowing argument pass-through (#128)
  • Improved package documentation (#127)

latrend v1.5.02022-11-10

New features

  • Implemented virtual class for meta methods (#61)
  • Implemented a meta method for repeatedly fitting a method and selecting the best fit (see lcFitRep, lcFitRepMin, lcFitRepMax) (#61)
  • Implemented a meta method for repeatedly fitting a method until convergence (see lcFitConverged) (#61)

Major changes

Other changes

  • Added check for formula argument to LMKM method
  • Added “converged” slot to lcModelPartition
  • Generic validate() checks for correct output length
  • Added workaround for erroneous R CMD check rmarkdown import note

latrend v1.4.32022-10-26

  • Removed usage of soon-to-be deprecated ggplot functions

latrend v1.4.22022-08-15

Bug fixes

CRAN

  • Resolved lcmm test error
  • Shortened runtime of vignettes
  • Shortened runtime of tests

latrend v1.4.12022-07-04

Bug fixes

  • Fixed bug in predict.lcModel() introduced by #116
  • Attempt to fix the documentation warning about how the escaped ampersand is not escaped

Breaking changes

  • Default initialization of lcMethodLcmmGMM and lcMethodLcmmGBTM is set to init = "lme" to support lcmm v2.0.0 #121

latrend v1.4.0

New features

  • Added RMSE and WRMSE metrics
  • plotMetric() now shows legend when grouping
  • externalMetric(lcModels) returns a named distance matrix

Bug fixes

  • Improved kml package compatibility; fixed errors on datasets with variable-length short trajectories
  • Resolved warnings from latrendBoot() and latrendCV()

Breaking changes

  • Removed redundant lcModelCustom in favor of lcModelPartition
  • Renamed lcMethodCustom to lcMethodFunction for clarity

latrend v1.3.02022-05-26

New features

  1. Parameterized method testing framework for lcMethod and lcModel implementations. See test.latrend().
  2. Warnings for missing data observations
  3. Better handling of data with missing observations

Breaking changes

  1. Removed support for the longclust package as it is no longer available on CRAN.

Other changes

  1. Updated examples, vignettes and tests to pass with _R_CHECK_DEPENDS_ONLY_ = true.

latrend v1.2.12022-01-06

  1. Fewer required imports
  2. Reduced test and example time
  3. Documentation of metrics and external metrics
  4. Enabled renv for CI

latrend v1.2.0

New features

  1. Greatly expanded documentation of lcMethod, lcModel, transformFitted(), and transformPredict().
  2. latrendBatch() evaluates and validates methods and datasets prior to any fitting (#49). This informs users of any errors as soon as possible. Moreover it makes it significantly easier to run parallel computations without the need to export parts of the global environment.
  3. Added seed argument to latrendBatch() (#47). Similar to latrendRep(), seeds are now generated for all methods, allowing for reproducible results.
  4. latrendBatch() now supports an expression for its "data" argument (#50).
  5. Variable argument pass-through for lcModel methods.
  6. Default implementation for predictForCluster().
  7. Added timing information to log output of latrend*() methods (#51).
  8. Added "unit" option to estimationTime().
  9. Implemented estimationTime() for lcModels.
  10. plot(lcModel) only shows trajectories when "what" argument is not specified.
  11. Support for lcModel objects without training data (#36).
  12. Made it easier to define new lcMethod subclasses by defining better default methods.
  13. plot() for lcModels (#48)
  14. latrend() and derivative methods automatically suppress console output when verbose = FALSE (#45)
  15. Better automatic axis breaks in metric plots (#44).
  16. trajectoryAssignments() signature that accepts a posterior probability matrix (#34)
  17. Added convenient mixture initialization options to lcMethodLcmmGBTM and lcMethodLcmmGMM based on standard (single cluster) linear mixed model fit.
  18. lcMethodRandom() accepts seed argument.
  19. Expand trajectory assignment input options for lcModelPartition.
  20. Methods can now be initialized by instantiating the S4 class using new() #56, #57. latrend() functions accept a character name of the method class.
  21. plotClusterTrajectories() to use logic of plotTrajectories() for showing trajectories #65.
  22. plotClusterTrajectories() with ribbon for trajectory range #68.
  23. logLik() for k-means based methods #70.
  24. Standard (interpolated) non-parametric cluster trajectory estimation through lcModelPartition and lcModelWeightedPartition #72.
  25. Option for disabling warning on redefining metrics and external metrics #75.
  26. Output warning in default lcModel postprob implementation #78.
  27. default predictAssignments() returns the assignments of trajectoryAssignments() when no newdata is specified #79.
  28. Lowered the number of required dependencies
  29. Implemented APPA and OCC metrics. #81.

Breaking changes

  1. Significant: trajectories() now returns the original training data, instead of the fitted (predicted) data (#32). This was done to improve clarity. Previous uses of trajectories() and plotTrajectories() should be replaced by fittedTrajectories() and plotFittedTrajectories(), respectively.
  2. Significant: Reworked lcMethod initialization to use the standard S4 mechanism #56.
  3. Minor: lcMethod implementations: prepareData() must now return an environment (#39). In the past, NULL was allowed, but this increased code complexity further down the process.
  4. Minor: estimationTime() is now an S4 generic method. This does not affect existing code.

Bug fixes

  1. Critical: Fixed predict() when cluster membership is specified for the new data (#40).
  2. Critical: Fixed computation of WMAE and WMSE metrics (#52).
  3. Fixed lcMethod argument evaluation for symbolic name input that equals the respective argument name (#41).
  4. Fixed strip() error related to use of eapply().
  5. Fixed output error for latrendBatch() when errorHandling = "pass" (#46)
  6. Defined estimation time for lcModelPartition and lcModelWeightedPartition (#38).
  7. Fixed default output of logLik.lcModel and other implementations (#37).
  8. Default metric() did not compute any metrics.
  9. Fixed indentation of messages for latrend() and derivative methods.
  10. Fixed computation of WMAE and WMSE metrics #52.
  11. sprintf warning when running latrendBoot() or latrendCV() #53.
  12. lcMethodLMKM cluster coefficients are wrong when standardization is enabled #69.
  13. lcMethodGCKM fails to fit for nClusters = 1 in latrendBatch() #71.
  14. fittedTrajectories() now uses output of fitted() instead of predict() #82.