Add LocationsWidget and plot_locations

[chrishalcrow]

Add a LocationsWidget and plot_locations, to complement the AmplitudesWidget and plot_amplitudes.

This plots the spike locations as a function of time, and is 99% the same as the plot_amplitudes function. You can control which direction you want to plot using e.g. directions_axis = 'x' and plot a histogram if you'd like.

An example:

import spikeinterface.full as si
_, drift_rec, sort = si.generate_drifting_recording(
    num_units=2, 
    duration=30,
    generate_displacement_vector_kwargs=dict(
        displacement_sampling_frequency=20.0,
        drift_start_um=[0, 50],
        drift_stop_um=[0, -50],
        drift_step_um=5,
        motion_list=[
            dict(
                drift_mode="zigzag",
                non_rigid_gradient=None,
                t_start_drift=0.0,
                t_end_drift=None,
                period_s=300,
            ),
        ],
    )
)

sorting_analyzer = si.create_sorting_analyzer(sorting=sort, recording=drift_rec)
sorting_analyzer.compute(['random_spikes', 'templates', 'spike_locations'])
si.plot_locations(sorting_analyzer, unit_ids=['0'])

Or

si.plot_locations(sorting_analyzer, unit_ids=['0'], plot_histograms=True)

Can plot multiple units, e.g.

si.plot_locations(sorting_analyzer, unit_ids=['0','1'])

Up for debate:

• is this a useful widget? I've been using it, so I think it is!
• is a histogram useful?
• Naming. We already have plot_spike_locations which plots the spike locations on the probe. This naming is designed to pair with plot_amplitudes.

[zm711]

I can dig this at least. I would check this out if we had it.

[samuelgarcia]

Good idea. I think it can be usefull.
Maybe we could move it to the same file than spike_location.

Another idea (not sure it is a good one) could be to move this logic in the spike_location widget using having options like:

• along_times=True/False
• with_histogram=True/False

[JoeZiminski]

Looks really nice! Under the hood this looks similar to the DriftRasterMapWidget as shown here but with the spikes coloured by unit membership rather than amplitude? In this case maybe there is scope to leverage the existing raster-plot backend and use the array of unit ids for colouring?

Incorporating this directly with DraftRasterMapWidget would be confusing but there might be scope for a common raster-plot backend (I don't think there is anything particularly drift-specific about the DriftRasterMapWidget, it is just spike depth x time) that these various functions could call, with different spike-colouring options. There is also a RasterWidget, I'm not sure the scope / intersection across these different widgets.

The histogram look really nice, just in case useful here is another version on this PR that I never got over the line.

[chrishalcrow]

Looks really nice! Under the hood this looks similar to the DriftRasterMapWidget as shown here but with the spikes coloured by unit membership rather than amplitude? In this case maybe there is scope to leverage the existing raster-plot backend and use the array of unit ids for colouring?

Incorporating this directly with DraftRasterMapWidget would be confusing but there might be scope for a common raster-plot backend (I don't think there is anything particularly drift-specific about the DriftRasterMapWidget, it is just spike depth x time) that these various functions could call, with different spike-colouring options. There is also a RasterWidget, I'm not sure the scope / intersection across these different widgets.

The histogram look really nice, just in case useful here is another version on this PR that I never got over the line.

That's a great idea, Joe! I just hacked together a 90%-the-same plot using DriftRasterMapWidget in a few lines. I think a common raster-plot backend is a good idea - I'll give it a go!

Thanks for the other feedback, all!