Fix get_traces for a local common reference

[chrishalcrow]

Fixes bug in get_traces for common reference recordings with a local reference, where you couldn't select a subset of channels. Noticed it when plotting traces.

Here's minimal code which reproduces the bug

from spikeinterface.core import generate_recording
from spikeinterface.preprocessing import common_reference

recording = generate_recording(durations=[5.0], num_channels=4)
recording_cmr_segment = common_reference(recording, reference="local")._recording_segments[0]

recording_cmr_segment.get_traces(start_frame=0, end_frame=10, channel_indices=[1,2])

This should raises an error.

Also added a test which selects a subset of channels ids, and makes sure it's selected the right one.

[alejoe91]

Hi Chris, does it fail if you call the recording.get_traces() as well? Or only the segment?

[chrishalcrow]

Hi Chris, does it fail if you call the recording.get_traces() as well? Or only the segment?

Morning! It works for get_traces on the recording. So it's only a problem if some functions (or users) use get_traces on the segments directly (e.g. in plot_traces).

[alejoe91]

@chrishalcrow instead of this "segment" test, I guess you can trigger the same issue by adding this line:

local_trace = recording_local_car.get_traces(channel_ids=["c", "d"])

after L71

I would actually prefer to extend that test instead of adding this new one, since it was designed (badly) to test the slicing behavior. Could you do that?

[chrishalcrow]

I have extended the test to include what you say, and also slicing the segments for each type of common reference. Also made the tests a little harder by choosing ["b", "d"] rather than ["a", "b"]. I also shorted the generated recording from 5s to 1s: I can't see any reason here not to do this and it doubles the speed of the test for me locally.

[alejoe91]

Thansk @chrishalcrow

Check comment!

[samuelgarcia]

even theses 2 lines looks very very bad. no ?

[alejoe91]

this is because channel_indices can be a slice and we need an iterator

[chrishalcrow]

Maybe they look bad, but they're not breaking anything! Is this ultimately due to the difference between channel_ids and channel_indices?