scikit-survival 0.18.0 released

I’m pleased to announce the release of scikit-survival 0.18.0, which adds support for scikit-learn 1.1.

In addition, this release adds the return_array argument to all models providing predict_survival_function and predict_cumulative_hazard_function. That means you can now choose, whether you want to have the survival (cumulative hazard function) automatically evaluated at the unique event times. This is particular useful for plotting. Previously, you would have to evaluate each survival function before plotting:

estimator = CoxPHSurvivalAnalysis(), y_train)

pred_surv = estimator.predict_survival_function(
times = pred_surv[0].x
for surv_func in pred_surv:
    plt.step(times, surv_func(times), where="post")

Now, you can pass return_array=True and directly get probabilities of the survival function:

estimator = CoxPHSurvivalAnalysis(), y_train)

pred_surv_probs = estimator.predict_survival_function(
  X_test, return_array=True
times = estimator.event_times_
for probs in pred_surv_probs:
    plt.step(times, probs, where="post")

Finally, support for Python 3.7 has been dropped and the minimal required version of the following dependencies are raised:

  • numpy 1.17.3
  • pandas 1.0.5
  • scikit-learn 1.1.0
  • scipy 1.3.2

For a full list of changes in scikit-survival 0.18.0, please see the release notes.


Pre-built conda packages are available for Linux, macOS (Intel), and Windows, either

via pip:

pip install scikit-survival

or via conda

 conda install -c sebp scikit-survival
Sebastian Pölsterl
AI Researcher

My research interests include machine learning for time-to-event analysis, causal inference and biomedical applications.