max_cluster_subset recommended values

[nikhilchandra]

Hi Jacob,

We noticed the addition of the new parameter "max_cluster_subset" which places an upper limit on the number of spikes used for building the clustering graph. Do you have any intuition yet as to what value(s) to set this parameter to?

Thanks, Nikhil

[jacobpennington]

You do not need to set it to anything. The current default, None, uses the existing clustering strategy.

We will likely set the default to be 25,000 in the future. That is unlikely to have a noticeable impact on most 2-3 hour (or shorter) recordings, since it would only have an effect if there is more than 500,000 spikes for a single grouping center (~40um probe section for Neuropixels) assuming cluster_downsampling=20 (the default). For longer recordings, this should significantly reduce runtimes.

[nikhilchandra]

When you say "2-3 recordings" do you mean "2-3 hour recordings"?

[jacobpennington]

Yes, that was a typo.