Competing-risks random survival forests for Python. 10-22× faster than rfSRC on real EHR data, scales to n=10⁶ in ~1 min, sklearn-compatible.
