big-life-lab/deeproc

deeproc's Issues

One rare case is not yet handled in the Partial C Statistic function for risk groups (Concordant Partial AUC is unaffected)

The Partial C Statistic function has a known issue for a case that has not yet been handled (coded). Time is the only limitation. The unhandled case (which is rare) is when the top and bottom of a group or partial curve, are within the same instance (see concordance matrix) then the computed value is incorrect. Usually this is a small value in absolute terms, with a correspondingly small error in absolute terms.

Some discrete measures do not interpolate the boundaries for risk groups

Some discrete measures such as avgPPV, avgNPV, avgLRp and avgLRn do not interpolate the group boundaries, but a warning is given that the closest boundaries matching instances are used instead. This affects risk groups, but it does NOT affect demographic groups (which are not interpolated). There is the option to choose the closest boundaries to those specified which avoids this warning and issue. Otherwise, non-interpolated measures will not match interpolated measures. This will be fixed with time.

Optimal ROC point in Mean ROC plots may not use a correct (weighted average) prevalence per labels (check)

The PyPI package deeproc 0.0.3 is not quite working yet (but the code downloaded directly from Github works)