set_probegroup fails when recording NWB file has sampling rate attribute instead of timestamps

[dervinism]

set_probegroup fails when recording NWB file has sampling rate attribute instead of timestamps, e.g.

recording = se.NwbRecordingExtractor(file_path=data_path, electrical_series_path='acquisition/TimeSeries_32000_Hz', \
                                     load_time_vector=True, use_pynwb=True, \
                                     cache=True, stream_cache_path=cache_path)
recording = recording.set_probegroup(probegroup)

It works fine when timestamps are explicitly specified when the recording NWB file is generated.

This is a shortcoming that should be fixed as NWB files can have either specification.

A repost from SpikeInterface/probeinterface#345 (comment)

[zm711]

@h-mayorquin, I trust your knowledge of nwb over mine.

[h-mayorquin]

Can you share your error trace?

[dervinism]

Here it is

Exception has occurred: UnboundLocalError
cannot access local variable 'timestamps' where it is not associated with a value
  File "/mnt/c/Users/m329786/code_repositories/nwb2spikesort/master_file.py", line 69, in <module>
    recording = recording.set_probegroup(probegroup)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
UnboundLocalError: cannot access local variable 'timestamps' where it is not associated with a value

[h-mayorquin]

Can you post the full one? It does not matter if it is very large.

[dervinism]

Sorry, this is it

Traceback (most recent call last):
  File "/mnt/c/Users/m329786/code_repositories/nwb2spikesort/master_file.py", line 69, in <module>
    recording = recording.set_probegroup(probegroup)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/m329786/anaconda3/envs/si_env/lib/python3.11/site-packages/spikeinterface/core/baserecordingsnippets.py", line 103, in set_probegroup
    return self._set_probes(probegroup, group_mode=group_mode, in_place=in_place)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/m329786/anaconda3/envs/si_env/lib/python3.11/site-packages/spikeinterface/core/baserecordingsnippets.py", line 194, in _set_probes
    sub_recording = self.clone()
                    ^^^^^^^^^^^^
  File "/home/m329786/anaconda3/envs/si_env/lib/python3.11/site-packages/spikeinterface/core/base.py", line 611, in clone
    clone = BaseExtractor.from_dict(d)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/m329786/anaconda3/envs/si_env/lib/python3.11/site-packages/spikeinterface/core/base.py", line 571, in from_dict
    extractor = _load_extractor_from_dict(dictionary)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/m329786/anaconda3/envs/si_env/lib/python3.11/site-packages/spikeinterface/core/base.py", line 1127, in _load_extractor_from_dict
    extractor = extractor_class(**new_kwargs)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/m329786/anaconda3/envs/si_env/lib/python3.11/site-packages/spikeinterface/extractors/nwbextractors.py", line 590, in __init__
    ) = self._fetch_recording_segment_info_backend(file, cache, load_time_vector, samples_for_rate_estimation)
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/m329786/anaconda3/envs/si_env/lib/python3.11/site-packages/spikeinterface/extractors/nwbextractors.py", line 767, in _fetch_recording_segment_info_backend
    if load_time_vector and timestamps is not None:
                            ^^^^^^^^^^
UnboundLocalError: cannot access local variable 'timestamps' where it is not associated with a value

[h-mayorquin]

Ah, @pauladkisson already found the cause. We need to discuss the solution. Let's do this next meeting.

[dervinism]

Have you found the solution? I would like to process my files as they are without resorting to regenerating them all with explicit timestamps.

[h-mayorquin]

We will discuss this tomorrow. Meanwhile, I have a question: can you tell me why are you setting a probe to NWB?

[dervinism]

Not sure if I understand correctly. I'm providing the probe info, so that channel coordinates are known. Isn't this info critical for spike sorting?

[dervinism]

Or does the spike sorter infers coordinates from the NWB electrode table anyway?

[h-mayorquin]

Yes, to better re-sate my question.

If your NWB has rel_x, rel_y and rel_z in your electrodes table those values should be loaded as locations which is what you need to sort data.

Are you working on a file without that info on the electrodes table?

[dervinism]

The table has columns x, y, and z. Would that work?

[h-mayorquin]

When they are stored as rel_x, rel_y and rel_z they are loaded automatically as locations (as they represent the relative geometry of the electrodes) but it seems to me that the absolute location should work as well?

So, do your x, y and z capture the internal geometry of the probe? If yes, you can set the locations yourself which is what is actually used for the sorting.

See, you need to do something like this:

recording.set_property("location", locations)

Which would be what set_probe does:

https://github.com/h-mayorquin/spikeinterface/blob/966cce8e65114f53e9b9eed8365af29adb480dc2/src/spikeinterface/core/baserecordingsnippets.py#L205-L211

Maybe we should load x, y and z as locations in the NWBRecordingExtractor if rel_x rel_y and rel_z are not found.

Let me know if this is clear