Some spike sorting methods can handle drifting on the waveforms\cite{Chung_2017,Jun_2017,Niediek_2016} . But if two neurons have a very similar waveform at different times they can not isolate them because they detach the waveforms from the timestamp.
Other methods use a gradual adaptation of a fixed number of neurons \cite{Franke_2009,Calabrese_2011,Pouzat_2004}. This allow them to track neurons even when the are not isolated, but they are sensitive to errors produced in the spike sorting algorithm used to detecte the number of classes, and dynamic changes in that number.
Furthermore, Spikes_Link has a lower computational cost than other alternatives \cite{Dhawale_2017,Bar_Hillel_2006}
It could be possible to use Spikes_Link in conjunction of SpikeInterface \cite{Buccino_2019}, a framework that unify multiple spike sorting methods. First a comparison of the results in a single block could determinate with spike sorting algorithm is the best for a given type of record and second if an interface between both frameworks is deevloped all the methods available on SpikeInterface will can be apply on Spikes_Link .
Finally, the framework presented in this work can be apply for other problems and it can be specially useful for data streams with concept drift and class imbalance \cite{Hoens_2012}.