optimizing main sorting parameters

[DanEgert]

Is there a good way to find better fitting main parameters (ops.Th, ops.AUCsplit and ops.lam)?
I have had unexpected strong variation in number of clusters in similar recordings and many merged clusters.
It would be cool to have a cluster metrics assessment that allows comparing the sort quality.

[marius10p]

A single set of parameters should work well for all data from the same probe and prep. The default set of parameters should work well in most cases.

If this is not the case, the underlying cause might be something about the data. Use the GUI to look at different sessions, especially the whitened image view, which is the preprocessing that Kilosort uses.

It is also possible that KS3 is more sensitive, and you can try using KS2.5.

Have you tried the "good" flag in the output to get a sense of how many good clusters you get?