bug with outliers
Opened this issue · 0 comments
braniii commented
In the publication, only datasets with sufficient samples and without outliers were considered. Thanks to Nicolò, it has been shown that this method can lead to negative entropies if neither is the case. The reason is that in case of significant outliers, e.g. caused by log-distributed data analyzed on a linear scale, the use of np.mean
on the radii is no longer meaningful and the volume is heavily overestimated. To work around this, one should implement an alternative with e.g. np.median
.
- Allow user to use the median for rescaling the radii.