Extend artifact detection: DetectAndRemoveArtifacts preprocessor and signed saturation

[alejoe91]

This PR introduces a DetectAndRemoveArtifactsRecording class so that it can be used in the preprocessing pipeline.
Users can pass a recording_to_detect: in this case, saturation/artifacts are detected on that recording (e.g., raw) and applied to the preprocessor parent (preprocessed)

Also extended saturation to optionally provide the "sign" of the saturation events. This can be useful to further inspect saturation issues and track down experimental causes to the saturation.

[chrishalcrow]

Hello, this is great. Part of (my) pipeline philosophy is that you can load a pipeline from a dumped json/pickle:

spikeinterface/src/spikeinterface/preprocessing/pipeline.py

Line 252 in a3f8db3

def _make_pipeline_dict_from_recording_dict(recording_dict):

I don't know how this will work when one argument is a recording object? Might just work?

And I also like the idea that you can specify a pipeline externally without reference to the recording, which this breaks. But more philosophical than practical... I can't really see a way around this.

[alejoe91]

Hello, this is great. Part of (my) pipeline philosophy is that you can load a pipeline from a dumped json/pickle:

spikeinterface/src/spikeinterface/preprocessing/pipeline.py

Line 252 in a3f8db3

def _make_pipeline_dict_from_recording_dict(recording_dict):

I don't know how this will work when one argument is a recording object? Might just work?

And I also like the idea that you can specify a pipeline externally without reference to the recording, which this breaks. But more philosophical than practical... I can't really see a way around this.

I have a suggestion. We can have some "magic" string:

• "pipeline['raw']": tells the function to use the input recording
• "pipeline['{preprocessing_step}']: grabs recording at a specific step.

What do you think?

[oliche]

It's good to go, maybe we could state that this apodization is a number of samples ?

Thanks a lot for looking at this !

[alejoe91]

It's good to go, maybe we could state that this apodization is a number of samples ?

Thanks a lot for looking at this !

Renamed apodization_samples ;)

[chrishalcrow]

So the idea of the magic string is to allow this to be

Suggested change

	recording_to_detect: BaseRecording | None = None,
	recording_to_detect: BaseRecording | str | None = None,

?
So that a user could pass a preprocessing pipeline

{
    'common_reference': {},
    'bandpass_filter': {},
    'detect_and_remove_artifacts': {'recording_to_detect': 'raw'}
}

or even

{
    'phase_shift': {},
    'common_reference': {},
    'bandpass_filter': {},
    'detect_and_remove_artifacts': {'recording_to_detect': 'phase_shift'}
}

When I type it out like this, it looks reasonable. But it only works if you're using a pipeline, since:

phase_shift_recording = si.phase_shift(raw)
remove_artifacts_recording = si.dectect_and_remove_artifacts(phase_shift_recording, recording_to_detect = 'phase_shift')

makes no sense.

Maybe we could add a private kwarg _using_pipeline, and you can only use a string if this is set to true? Or maybe the Pipeline object can be responsible for converting the string into a recording.

Or maybe right now we just add the "raw" option?