I corrected and improved the ES implementation during my bachelor thesis at PIK. The issues were discussed with and confirmed by Reik Donner as well.
Two main issues:
Normalisation was carried out incorrectly. The normalisation must take into account that the first and last event in each time series is only used to calculate tau, but does not contribute to the counting function. Therefore two events must be subtracted for the normalisation in each time series.
The code used some redundant calculations in large arrays, which was unfavourable. This implementation runs approximately twice as fast.
I corrected and improved the ES implementation during my bachelor thesis at PIK. The issues were discussed with and confirmed by Reik Donner as well.
Two main issues: