Concordance index errors with repeated events

The following produces a nan value in lifelines version 0.9.1.0.

>>> import lifelines
>>> import numpy as np
>>> from lifelines.utils import concordance_index
>>> times = np.arange(3)
>>> events = np.ones(3)
>>> concordance_index(events, times)
nan

I don’t think that’s the expected outcome, it should be 0.5, right?

The same error happens when n=1. In this case, I think a ValueError should be thrown for input validation though.

>>> import lifelines
>>> import numpy as np
>>> from lifelines.utils import concordance_index
>>> times = np.arange(1)
>>> events = np.ones(1)
>>> concordance_index(events, times)
nan

[AyuranS]

@CamDavidsonPilon , I realised on trying to reproduce the bug, the issue was due to having some NaNs in the actual Time To Event column when the input is a DataFrame. This doesn’t really make it a use case for the function anymore but maybe the input error isn’t being handled correctly. I believe this would reproduce the error at least on my machine. `test1= pd.DataFrame(np.random.randint(800, size=800), columns=[‘prediction’])

test2= pd.DataFrame(np.random.randint(800, size=800), columns=[‘actuals’])

test2[-50:-1]=np.nan

print(test1.shape)

print(test2.shape)

print(lf.utils.concordance_index(test2, test1))`

[CamDavidsonPilon]

So I’ve added an ZeroDivisionError to the latest version of pypi, 0.9.2. I’ve left the c-index as is, according to the second paper there.