Skip to contents

Changelog

Source: NEWS.md

targeted 0.9.0.9000

  • learnerSL: new R6 class for super learners, which inherits from learner. learnerSL$new() has the same arguments as learner$new(), where the base learners are provided via estimate.args$learners. learner_sl now returns a learnerSL object.

Bug fixes:

  • learner$clone() produced a clone with unintended reference-type behaviour. A clone used the formula of the original object, instead of a new formula that was provided with the update() method.
  • learner_sl: update now changes the response variable of all base learners, while each base learner keeps its covariates. Previously, only the formula of the super learner was updated, such that, e.g., cate estimated treatment models of the super learner with the wrong response variable.
  • learner$update() with a character string lost the environment of the formula. Formula variables that are not in the data (e.g., defined inside a function) were not found or resolved to other objects with the same name (e.g., cut to base::cut()). A new response variable now keeps the environment of the current formula, and a formula string gets the calling environment like a formula object. This also fixes aipw(), cate() and moi(), which update the response variable of the provided learners.
  • design() now throws an error for unnamed arguments in ... and for abbreviations of subset (e.g., sub =). Such arguments were passed on to the subset argument of model.frame(), which silently dropped or duplicated observations (#241).

targeted 0.9

  • cate: doubly robust estimation with missing outcomes under MAR via the new missing.model argument
  • summary.cate reduced object size

targeted 0.8

CRAN release: 2026-07-15

  • moi: implementation of one-step outcome imputation estimator (arXiv.2606.07174):
  • design tolerates now special variables who are absent in data
  • rename rm_envir argument of design to rm.envir
  • adding predict_filter generator functions for experimental predict.filter argument of learner
  • removing deprecated ml_model
  • learner_surv_cox, learner_surv_rf, cumhaz
  • feature(superlearner): more robust metalearner (in case quadratic prog. fails)
  • feature(learner_cox): mets::phreg wrapper.
  • disabling openMP in RcppArmadillo
  • bugfix: issue with cate with repeated cross-fitting fixed.

Breaking changes:

  • removing legacy crr and cate_link functions
  • removing legacy SL function. Use learner_sl instead.

targeted 0.7.1

CRAN release: 2026-01-12

Maintenance release

targeted 0.7

CRAN release: 2025-12-10

This release improves the multiple testing procedures

  • test_zmax_onesided: one-sided Zmax / minP test
  • test_intersection_sw: the following parameters for the optimization can be controlled via the control list argument: dykstra_niter sets the maximum number of iterations (default 500), dykstra_tol convergence tolerance of the alternating projection algorithm (default 1e-7), pinv_tol tolerance for calculating the pseudo-inverse matrix (default value: nrow(vcov).Machine$double.epsmax(eigenvalue)).
  • test_intersection_sw handle edge-case where vcov is not positive-definite. Throw warning and project onto nearest PD matrix in Frobenius norm.

Breaking changes:

  • design refactored. Better handling of response variables (factors) xlev argument changed to levels.
  • learner_xgboost updated to reflect changes in xgboost 3.1.2.1. Note, the arguments names have changed accordingly and introduces thus possible breaking changes.

Features

  • (cate) tibble df and data.table are converted to data.frame (#135) - (905bcf1)
  • (design) refactoring and better handling of formulas and response variables (#136) - (044e817)
  • (superlearner) standard meta-learner based on quadprog::solve.QP - (4c830b7)
  • (test_intersection_sw) unit tests (#137) - (10a793b)
  • (test_intersection_sw) add user-exposed optimization arguments and improve documentation (#134) - (4c406df)
  • (test_intersection_sw) constrained least squares via Dykstra’s algorithm, and fast signed Wald test evaluation - (2feaa87)
  • Feature/zmax (#129): onesided Zmax / minP test and clean-up of test_* + summary methods (#130) - (85cd87e)

Bug Fixes

  • (cate) Linear calibration (8dd4ca2)
  • (design) Print method handling Surv + factor objects (#138) - (6ce8c8a)
  • (makefile) Using remotes instead of devtools, to prevent workflows from breaking (#131) - (32d6b59)
  • (makefile) Using remotes instead of devtools, to prevent workflows from breaking - (bc4c6a6)
  • (slowtest) Metalearners needs to be referenced with ::: (not exported yet) (#132) - (e2324bb)
  • (slowtest) Metalearners not exported - (147740a)
  • (learner) Fix data.table and xgboost issues - (6bd9949)
  • (cate) Fix bug in cate: combn - (8a1dff9)
  • (aipw) Fixes issues with new cate - (82ac974)

Developer

  • (cate) Fix partial argument matching in cate_est call (#126) - (e12efe9)
  • (unscoped) docs(vignettes): moving to quarto documentation (#140) - (b896eba)

targeted 0.6

CRAN release: 2025-10-30

This release introduces a new learner class replacing the previous ML constructor.

  • constructors for commonly used regression and classification models are also implemented: learner_glm, learner_gam, learner_grf, learner_hal, learner_glmnet_cv, learner_svm, learner_xgboost, learner_mars, learner_isoreg, learner_naivebayes
  • new ensemble models (super-learners) available with superlearner and learner_sl
  • learner_stratify: implementation of learner that can stratifies base-learner on categorical predictor
  • learner_expand_grid: utility function to construct learners

Improved implementation of cate with repeated cross-fitting via the new ‘rep’ argument. Linear calibration via the calibration.model argument doi:10.1093/biomet/asaf029.

Implementation of estimators for joint modelling of time-to-event (CIF) and clinical outcome truncated by competing risk (arXiv.2502.03942): estimate_truncatedscore.

test_intersection_sw Constrained least squares via Dykstra’s algorithm, and fast signed wald test evaluation.

Features

  • (test_sw) signed wald intersection test - (ba58a2d)
  • (cate) linear calibration (#119) - (05a299f)
  • (cv) cross-validation cv method for superlearner objects (#64) - (1d58b26)
  • (design) Fixing how specials is handled and passed to learner functions - (ab46749)
  • (design) Adding print.design (#94) - (20eb170)
  • (learner) learner_stratify implementation of learner that can stratifies base-learner on categorical predictor (d561ea1)
  • (learner) [breaking] changing formula public field to active binding (#98) - (1505453)
  • (learner) [breaking] removing response.arg and x.arg arguments from learner$new() (#92) - (4043dd7)
  • (learner) adding new summary method to provide more details than print method (#87) - (d12a581)
  • (learner) changed behaviour of learner$design to return not only ‘x’ matrix but everything including ‘specials’ (#76) - (ca74abb)
  • (superlearner) new ensemble models (super-learners) (#104)
  • (learner) learner_expand_grid utility function to construct learners (#96) - (3ae461a)
  • Generalized Additive Models learner_gam (#77) - (de2ec2b)
  • Highly Adaptive Lasso learner_hal (#75) - (62c4941)
  • Elastic net learner_glmnet_cv (#74) - (67ba241)
  • Generalized Linear Models learner_glm (#63) - (0d2663a)
  • Naive Bayes classifier learner_naivebayes (#88) - (2cbe979)
  • Generalized Random Forest learner_grf (#84) - (82f76c8)
  • Support Vector Regression learner_svm (#83) - (4b28b30)
  • Isotonic regression learner_isoreg (#82) - (e409b58)
  • XGBoost learner_xgboost (#80) - (72ee414)
  • Multivariate Adaptive Regression Splines learner_mars (#79) - (0019060)
  • Super-Learner learner_sl (#78) - (03a81d2)
  • Adding new learner R6 class to replace ml_model (#68) - (86c44fd)
  • Improved riskreg_cens estimator (#62) - (7aef75f)
  • truncatedscore default is now to estimate P(T>=t) instead of CIF (#46) (b315645)
  • (cv) silent arg (#34) - (bb3d782)
  • Feature/truncatedscore Implementation of estimators for joint modelling of time-to-event (CIF) and clinical outcome truncated by competing risk. (#13)

Documentation

  • (learner) Harmonize documentation of all learner constructors (#93) - (1d37075)
  • (learner) New vignette on prediction model class (#4) - (08bedd4)
  • (RATE) Fix linter and re-use roxygen documentation (#45) - (a31a37c)

Bug Fixes

  • (cate) Now sets correct mc.cores argument in mclapply (#24) - (260ded8)

Developer

  • Adding add_dots utility function (#2) - (bb21da4)
  • Adding .cliff.toml (#47) - (47d7038)
  • Github workflows (#33) - (bcd50bd)
  • Adding custom function to inform users about deprecated function arguments (#32) - (d0865a2)
  • Makefile + repository cleaning (#23) - (fa39827)
  • Switch from testthat to tinytest for unit testing of R package (#6) - (be86072)
  • Adding .lintr config for R code linter - (7fe7b56)

targeted 0.5

CRAN release: 2024-02-22

  • cate now also returns the expected potential outcomes and influence functions
  • Bug-fix in the ml_model$update() method
  • The default scoring method for cv now only switches to log-score+brier score when the response is a factor. Custom model-scoring function (cv argument modelscore) automatically gets ‘weights’ appended to the formal-arguments.

targeted 0.4

CRAN release: 2023-12-19

  • alean: Assumption Lean inference for generalized linear model parameters
  • ate now supports general family argument
  • cate now supports parallelization via the future or parallel package
  • ml_model refactored. ML new wrapper for various machine learning models.
  • cv parallelization (future or parallel package)
  • riskreg_cens cumulative risk, restricted mean survival predictions (censored unbiased regression estimates)

targeted 0.3

CRAN release: 2022-10-25

  • Conditional average treatment estimator cate, crr
  • Generic prediction model class ml_model
  • design
  • SuperLearner wrapper SL
  • Average Treatment among responders RATE

targeted 0.2

  • Weighted Naive Bayes classifer with NB
  • Pooled adjacent violator algorithm pava
  • ODE solver ode_solve
  • Calibration calibration
  • Cross-validation cv
  • ace method updated and renamed to ate

targeted 0.1.1

CRAN release: 2020-05-15

  • Maintenance release.

targeted 0.1

CRAN release: 2020-05-08

  • Initialization of the new package targeted with implementation of augmented inverse probability weighting methods for estimation with missing data and causal inference (aipw, ace), and double robust methods for risk regression with binary exposure variables (riskreg).