Kilosort3 - Big differences in number of good units found

[Kasia-Bz]

Hello, I am doing chronic recordings with Neuropixel probes in rats. I am using Kilosort3 for spike sorting and I get a lot of varying outputs. I sorted one data set and ended up with 412 good units found, but from the same animal from recordings done two days earlier I got only 170 good units found. From other animal I get 169 units from one recording session and then only 29 from different recording session and 272 from a third day. These sessions are of the same duration, the same behavioural task and the same room, so it is difficult for me to find a reason for such a big differences. Is there any explanation and solution for it?

[DanEgert]

Any progress on this? I've seen similar issues. Sorting data from a 256ch oversampling array recorded on days 28-30. I don't expect losing a lot of clusters in this period. KS3 went from 250, 160 to 120. KS2.5 from 192, 120 to 97. It helped a bit to turn off bad channels, but I still see much larger variation than it should be.

[aeazraoui]

Same here. I run the same parameters on the same recording trace and get some variability in the output (+/- 5 good units).