Initial dartsort wrapper implementation.

[samuelgarcia]

@cwindolf @julienboussard @colehurwitz

[samuelgarcia]

@cwindolf any news ?

[chrishalcrow]

Suggested change

	# The main channel ids are kept in the Templates, which are saved in the sorter_output_folder
	matching_model_folders = list(Path(sorter_output_folder).glob("matching*_models"))
	if len(matching_model_folders) > 0:
	matching_model_folders.sort(key=lambda p: int(p.stem[8:-7]))
	last_matching_model_folder = matching_model_folders[-1]
	if (template_data_file := last_matching_model_folder / 'template_data.npz').is_file():
	from dartsort.templates.templates import TemplateData
	template_data = TemplateData.from_npz(template_data_file)
	main_channel_indices = template_data.main_channels()
	main_channel_ids = recording.channel_ids[main_channel_indices]
	sorting.set_property('main_channel_id', main_channel_ids)

We can add main_channel_ids by taking them from the templates file (as suggested by @cwindolf). Am testing on some real data now...

[chrishalcrow]

Sorry, this doesn't work. A unit can be made from spikes from several templates, so there's no one to one map from the templates to unit main_channel_ids.

[alejoe91]

but all templates associated to a unit should roughly have the same channels no? We could take the mode of the main channels from the templates associated to the same unit

[alejoe91]

@cwindolf could this be implemented directly in dartsort? so that the to_numpy_sorting will already have this property

[chrishalcrow]

spike_channels = dartsort_sorting.channels
main_channel_indices = [mode(spike_channels[labels == unit_id])[0] for unit_id in sorting.unit_ids]
main_channel_ids = recording.channel_ids[main_channel_indices]
sorting.set_property('main_channel_id', main_channel_ids)

Dartsort keeps the channel of each spike! So we can easily take the mode of the spike channels for each unit.

[chrishalcrow]

This works!!

[alejoe91]

Ok fixed a couple of stuff:

• using np.bincount(...).argmax() for mode
• saving to spikeinterface folder instead of NumpySortingFolder.write_sorting to preserve the added main_channel_id property (this should be equivalent after Move the Base.save to Sorting.save() and Recording.save() #4671)

[alejoe91]

Ok to merge?

[cwindolf]

Looks good to me! Thank you guys. Just adding some website links basically here.