BUG: cluster_templates function returns vastly different waveforms from the one in `templates.npy

[teristam]

Describe the issue:

Hi,

I try to get the template waveform of a cluster both from the new cluster_templates function and the original templates.npy. I notice the waveform returned by both are vastly different from each others:

I understand that the results from cluster_templates are scaled according to the spike amplitude, but even the best channel calculated within the cluster_templates() is different from the results from get_best_channels()

Upon looking closer, I notice that in

Kilosort/kilosort/data_tools.py

Lines 149 to 150 in db107d7

	spike_idx = spike_clusters[spike_clusters == cluster_id]
	temps = get_templates(spike_idx, results_dir)

The spike_idx refers to a list of cluster id, but in

Kilosort/kilosort/data_tools.py

Line 185 in db107d7

template_idx = spike_templates[spike_idx]

spike_idx actually refers to the index of detected spikes.

Is there a potential issue there?

Reproduce the bug:

Error message:

Version information:

4.0.27

[jacobpennington]

Thanks for catching that, there was indeed a mixup there. It's fixed with the latest changes, you can pull those now if you installed from source. It might be a day or two before they're available through pypi.

[teristam]

Thank you for looking into this!

However, I notice now that the result from cluster_templates is no longer scaled according to the amplitude of each spikes, different from what the doc of the function is saying.

The 'template' is scaled separately for each
spike based on its amplitude.

Is this the expected behaviour?

[jacobpennington]

How are you determining they are "no longer scaled?" They are scaled based on the values in amplitudes.npy, in data_tools.get_templates (line 102):

scaled = amplitudes[spike_idx, np.newaxis, np.newaxis] * temps

[teristam]

Sorry, my bad. I got confused about the get_templates and cluster_templates functions.

[teristam]

Hi,

I am re-opening this issue because my current problem seems to be related. I am using the cluster templates as a way to measure how close the spikes of a cluster are to the centroid of the cluster. My understanding is that the cluster templates are scaled in a way that should have similar amplitudes to the original spikes. However, when I plot the raw/mean spike waveforms together with the cluster template, they seem to have very different amplitudes. I got my spikes from get_spike_waveforms.:

Even in the original example in the doc (https://kilosort.readthedocs.io/en/latest/tutorials/plotting_example.html), the mean spike and template seems to be very different from each other in amplitude?

What is the best way to get a spike template that has similar amplitude to individual spikes?