LoginSign Up
← Back to all sparks
M

mlr3cmprsk

ANALYTICS
Velocity0.0

Competing Risks Machine Learning for mlr3

mlr3cmprsk.mlr-org.com ↗

Competing risks arrive in mlr3, going from a non-parametric baseline to Fine-Gray regression in seven weeks.

mlr3competing-riskssurvival-analysismachine-learningcumulative-incidence
◆Current state
mlr3cmprsk extends mlr3 to competing-risks analysis: learners predicting cumulative incidence functions, plus measures to score them. It appeared in February 2026 and reached 0.0.5 by April, adding a Fine-Gray learner and a Brier score along the way. The version numbers say pre-release; the cadence says active build-out.
◆Where it's heading
The package assembled a working evaluation stack fast. It started with the Aalen-Johansen estimator as a non-parametric baseline, added Fine-Gray as a regression alternative, then filled in the scoring side with an AUC refactor and a fixed-time Brier score. Development has run in lockstep with survdistr and mlr3extralearners, which picked up the dependency at 1.5.2.
◆Prediction
The visible gaps are more learners — cause-specific Cox and boosting variants — and the tuning and pipeline integration the rest of mlr3 expects; the measure side currently looks further along than the model side.

◆Recent moves

  1. 4mo ago

    Brier score for competing risks and an AUC refactor

    msr("cmprsk.brier") adds fixed-time-point prediction error evaluation, and MeasureCompRisksAUC gains a cause_weights parameter with time_horizon renamed to time. Test coverage passes 80%.

    View source ↗
  2. 4mo ago

    Fine-Gray learner brings regression modelling to competing risks

    ⚡ SPARK

    LearnerCompRisksFineGray arrives via the cmprsk package, giving the framework a subdistribution-hazard regression model alongside the non-parametric Aalen-Johansen estimator it launched with. merge_cifs() is also renamed to align_cifs().

    View source ↗
  3. 4mo ago

    Aalen-Johansen learner gains standard mlr3 properties

    Adds missings, importance and selected_features properties plus native_model() support to the Aalen-Johansen learner, and moves its CIF predictions to unique event times. The mlr3 floor rises to 1.4.0.

    View source ↗
  4. 4mo ago

    CIF interpolation pulled from survdistr

    Picks up an interpolation function from survdistr 0.0.2 for CIF curves; internal plumbing with no interface change.

    View source ↗
  5. 5mo ago

    First release on mlr-org.r-universe

    The package's initial deployment on mlr-org.r-universe, establishing competing risks as a task type within the mlr3 ecosystem.

    View source ↗