Question MergeUnitSorting unit_ids

[Antho2422]

Hi,

I have a question about the merge performed using MergeUnitSorting.

If sorting.get_unit_ids() = [0, 1, 2, 3] and I merge unit 0 and 1 together the output will be new_sorting.get_unit_ids() = [2, 3, 4] right ?

Is it always the case and if not, is there a way to make sure of which merge resulted in which unit ?

I want to make sure that I'm looking to the right units when a merge is performed.

Thanks,
Anthony

[Antho2422]

I also noticed some weird things when plotting the templates of the units I merge and the merged result. It seems like the waveforms is shifted on the right, I don't know what could be the reason because the number of samples of the templates is the same.

Here is an example of what I mean (the red unit is, if the above is right, the merged unit):

[JoeZiminski]

Hi Anthony, it seems the default behaviour of merge is to use the next number than has never been used for an id before (see here) and here. If for example you merge 2,3 the new unit will be 5 (it will not go back to 0 or 1). However, you can overwrite this behaviour by specifying new_unit_id (see end of the example below). Someone else with a bit more familiarity with this might be able to weight in though.

That is interesting about the merged plots, could you please post the code you are using (or a minimal reproducable example)? The test code below seems to look okay:

import spikeinterface as si
import matplotlib.pyplot as plt
import spikeinterface.curation as si_curation
import numpy as np

# Generate curated sorting
recording, sorting = si.generate_ground_truth_recording()
curation_sorting = cs = si_curation.CurationSorting(parent_sorting=sorting)

# first extract the templates for units 0 and 1
waveforms = si.extract_waveforms(recording, curation_sorting.sorting, mode="memory")

print(waveforms.sorting.get_unit_ids())
# [0 1 2 3 4 5 6 7 8 9]

unit_0 = waveforms.get_template(unit_id=0)
unit_1 = waveforms.get_template(unit_id=1)

# Then merge zero and one, and extract the new merged template (unit 10)
curation_sorting = si_curation.CurationSorting(parent_sorting=waveforms.sorting)
curation_sorting.merge(units_to_merge=[0, 1])
print(curation_sorting.sorting.get_unit_ids())
# [ 2  3  4  5  6  7  8  9 10]

curated_waveforms = si.extract_waveforms(recording, curation_sorting.sorting, mode="memory")
merged_0_and_1 = curated_waveforms.get_template(unit_id=10)

# Plot all templates
plt.plot(np.average(unit_0, axis=1))
plt.plot(np.average(unit_1, axis=1))
plt.plot(np.average(merged_0_and_1, axis=1))
plt.legend(["0", "1", "merged"])

plt.show()

# Play around with merging to check the behaviour
curation_sorting.merge(units_to_merge=[2,3])
print(curation_sorting.sorting.get_unit_ids())
# [ 4  5  6  7  8  9 10 11]

curation_sorting.merge(units_to_merge=[4,5], new_unit_id=0)
print(curation_sorting.sorting.get_unit_ids())
# [ 6  7  8  9 10 11  0]

[Antho2422]

@JoeZiminski

Hey, thanks a lot for your answer. For the unit ids it is much more clear now thanks a lot.

About the plotting issue, I need to double check, this may come from my side. I will rework on that when I will have some time but seeing your example it seems to work fine so it may come from my code.

Thank you a lot,

Anthony

[zm711]

I'll close this one for now. But if you still see the problem just reopen.