NEWS

---

polle 1.6.7

• q_sl() and g_sl() are reinstated, with SuperLearner (>=2.0-42) in Suggests rather than Depends.

polle 1.6.6 (2026-10-10)

• bug fix: the risk set used by fit_c_functions() and predict.c_functions() now correctly excludes subjects whose terminal event has already occurred (events in {0, 2} at stages 1..K and {1, 2} at stage K+1).
  • bug fix: policy_eval(type = "dr")$coef_ipw is now correctly computed when some subjects terminate before stage K+1.
  • breaking: policy_eval(type = "ipw") and policy_eval(type = "or") are no longer supported. The accompanying inverse-probability-weighted and outcome-regression-only estimates remain available on a type = "dr" fit as coef_ipw and coef_or. Calling policy_eval() with type = "ipw" or type = "or" now raises an informative error pointing at the migration path.

polle 1.6.5 (2026-09-30)

• SuperLearner was removed from the package dependencies in anticipation of its CRAN archival; it has since been reinstated as Suggests, see the 1.6.6 entry above. q_sl() and g_sl() were temporarily unavailable in this release.
  • control_owl() gains a solver argument, defaulting to "ipop". DTRlearn2 (>= 2.1) changed the default weighted-SVM solver of DTRlearn2::owl() to "svm" (via WeightSVM::wsvm)
  • bug fix: final-outcome and discrete-censoring model fitting to complete after warnings, retaining the fitted models and preserving the learner warnings.
  • bug fix: set Q_{K+1} = U for early terminal events under missing final outcome

polle 1.6.4 (2026-05-17)

• new citation, doi:10.18637/jss.v116.i04, Journal of Statistical Software

polle 1.6.3 (2026-05-05)

• policy_def() gains natural_action and stage_number arguments
  • bug fix: c_cox() now passes time, time2 and event to mets::phreg

polle 1.6.1 (2025-12-01)

• policy learning and evaluation under right-censoring/missing outcome

polle 1.6.0 (2025-10-30)

• online estimation/sequential validation for (subgroup) policy evaluation
  • improved functionality for subgroup analysis (subgroup average treatment effect)

polle 1.5.1

• documentation and print method for policy_eval() improved
  • multiple thresholds for policy_learn()
  • bug fixes: get_policy_functions(), predict.blip_function(), predict.q_glmnet()

polle 1.5 (2024-09-06)

• added target argument in policy_eval for estimating the subgroup average treatment effect.
  • added threshold argument in policy_learn for learning the optimal subgroup.
  • added vignette for learning and evaluating the optimal subgroup.

polle 1.4 (2024-04-25)

• vignettes added for policy_data, policy_learn and policy_eval.

polle 1.3 (2023-07-06)

• new policy_learn type: "blip". Similar to "drql", but only a single model is fitted using the doubly robust score for the blip.
  • q_sl now uses the folds from policy_learn when it is used a policy model for "blip" and "drql".
  • g_xgboost and q_xgboost.

polle 1.2 (2023-02-07)

• action sets can now vary across stages (stage_action_sets)
  • g_empir() is a new g-model useful for calculating the empirical (conditional) probabilities
  • conditional() estimates the group specific policy value estimates
  • progressr is now implemented for policy_eval()
  • sim_single_stage(), sim_two_stage(), and sim_multi_stage() are new functions for simulating data.

polle 1.0 (2022-12-06)

• Package documentation: https://arxiv.org/abs/2212.02335