Sorting analyzer removes _recording from sorting object

[ecobost]

print(sorting.has_recording()) # True
analyzer = sa.create_sorting_analyzer(sorting, recording=sorting._recording)
print(analyzer.sorting.has_recording()) # False

Is this expected or a bug? Not sure why the recording would be unregistered.

I was doing something like curated = sc.remove_redundant_units(analyzer) and found curated suddenly does not have the recording anymore (because remove_redundant_units uses analyzer.sorting internally).

[alejoe91]

Good point! We should register the recording internally (if not in recording less mode!)

[ecobost]

When you create an analyzer (binary_folder), it creates the analyzer folder, saves recording info (as recording.json) and sorting object (as numpy_folder sorting/), and then reloads the analyzer folder to create the analyzer. The issue is the sorting (as NumpyFolderSorting) does not save info about its recording so when the analyzer is re-loaded (during create_sorting_analyzer->create->create_binary_folder->load_binary_folder), analyzer.sorting loses the link to its original recording.

spikeinterface/src/spikeinterface/core/sortinganalyzer.py

Lines 542 to 544 in 797f0e8

	sorting = NumpySorting.from_sorting(
	NumpyFolderSorting(folder / "sorting"), with_metadata=True, copy_spike_vector=True
	)

In the sorting I was looking at annotations.__sorting__info__.recording has recording info because this NumpyFoldersorting was generated from a kilosort object (which has recording as annotation) but the existence of this annotation is not guaranteed (I think).

The non-hacky solution will be to save the recording information when a Sorting object is saved (possibly as a spikeinterface_recording.json file as in the sorter_output/ folders or as recording.json as done in analyzer) and then load it when a NumpyFolderSorting object is initialized. That will also solve #4413