search

MouseLand / Kilosort

Fast spike sorting with drift correction
https://kilosort.readthedocs.io/en/latest/
GNU General Public License v3.0
477 stars 248 forks source link

same spike, member of two clusters #193

Closed SaraMati closed 4 years ago

SaraMati commented 4 years ago

I've observed sometimes the same spike has been associated to two clusters.

in the screenshot below, cluster 14 has the same spike as cluster 15. I've attached two screenshots with changing colors for clusters to make it easy to identify the middle spike that is in both clusters.
also, why cluster 14 has so many flat waves fitted to it?

I'd be grateful if you help me spot what is wrong here (any config adjustment to avoid this?)

image image

SaraMati commented 4 years ago

I would be much grateful if someone addresses the issue of same spikes being sorted into several clusters. I still see the issue, as the picture below in which the 4 spikes shown are sorted into clusters 54, 55, and 49.

image

has this issue been discussed anywhere? can you refer me to it?

agbondy commented 4 years ago

See the issue thread here: https://github.com/MouseLand/Kilosort2/issues/29

This is a well known issue that is a fairly fundamental consequence of the template matching approach used by Kilosort. The current solution is to remove double-counted spikes in post-processing, for which that thread provides some pointers.

SaraMati commented 4 years ago

thank you very much