0.20.1
- Some performance improvements to
CoxPHFitter(about 30%). I know it may seem silly, but we are now about the same or slighty faster than the Cox model in R'ssurvivalpackage (for some testing datasets and some configurations). This is a big deal, because 1) lifelines does more error checking prior, 2) R's cox model is written in C, and we are still pure Python/NumPy, 3) R's cox model has decades of development. - suppressed unimportant warnings
API changes
- Previously, lifelines always added a 0 row to
cph.baseline_hazard_, even if there were no event at this time. This is no longer the case. A 0 will still be added if there is a duration (observed or not) at 0 occurs however.