How to resolve NaNs and/or zeroes present in weights for spikedetect.run?

[NirvikNU]

Describe the issue:

I am using data from a 32 contact laminar probe data (150 micron spacing) in Kilosort4. The data is in int16 format. Kilosort3 was able to detect spikes and sort it successfully.

However, I repeatedly get this warning when running the same file with Kilosort 4: d:\kilosort\kilosort\kilosort\spikedetect.py:232: UserWarning: NaNs and/or zeroes present in weights for spikedetect.run, may need to adjust min_template_size and/or dminx for best results.

subsequent to this warning, the program terminates with the error: 0 spikes extracted in 33.26s; total 33.47s Traceback (most recent call last): File "d:\kilosort\kilosort\kilosort\gui\sorter.py", line 82, in run st, tF, Wall0, clu0 = detect_spikes(ops, self.device, bfile, tic0=tic0, File "d:\kilosort\kilosort\kilosort\run_kilosort.py", line 397, in detect_spikes raise ValueError('No spikes detected, cannot continue sorting.')

I have turned off drift correction, also tried playing with the parameters dmin, min_template_size, nearest_templates, but all the combinations I tried keep generating the same warning followed by the same error. Is there a systematic way to test the parameters to find the right combination?

[jacobpennington]

Hello,

Am I understanding correctly that this is 32 contacts along a single shank, i.e. all x-positions are the same? If so, I would try setting dmin = 300. Template centers are placed every dmin//2 microns, which allows for some overlap for densely spaced contacts, but in your case the default dmin = <median difference in y position> would center half of the templates in the large gaps between contacts.

I'm not sure what will work best for nearest_templates, but you will want something much smaller than the default. For linear probes, there will always be a single template center laterally. So, given how far apart the templates would be with dmin = 300, I would try nearest_templates = 3 so that only immediate neighbors are compared except at the ends. You could probably reduce nearest_chans to 3 as well for the same reason.

Increasing min_template_size to something like 20um might help with the NaN weight issue, but I would try the other parameters first.

[NirvikNU]

Thanks a lot! Changing nearest_templates = 3and dmin = 300 worked so as far now spikes are detected and the sorting is completed. However, I do still get the NaN issue. I tried increasing min_template_size to 20 but the warning still persists. Any suggestions as to why it still persists?

[jacobpennington]

@NirvikNU It happens when the distances between templates are really large compared to min_template_size. It doesn't necessarily need to be fixed, the warning is just there as a debugging tool in case no spikes are detected, and sometimes increasing min_template_size a bit will be enough to get rid of it. As long as the sorting finished and the results look reasonable (i.e. similar to your KS3 results), don't worry about it.

[NirvikNU]

I see. That makes sense. Thanks!