applying curaiton fails if all units were either split or deleted

[LizzyMcKay]

Hi,

I just noticed a possible bug. If all the units in a json curation file (created by spikeinterface-gui) are either split or deleted, applying curation fails:

IndexError                                Traceback (most recent call last)
Cell In[16], line 1
----> 1 analyzer_curated = scur.apply_curation(sorting_analyzer,curation_dict_or_model=cur)

File ~/arbeit/PyCode/spikeinterface/spikeinterface-git/src/spikeinterface/curation/curation_format.py:287, in apply_curation(sorting_or_analyzer, curation_dict_or_model, censor_ms, new_id_strategy, merging_mode, sparsity_overlap, raise_error_if_overlap_fails, verbose, **job_kwargs)
    279         curated_sorting_or_analyzer, _ = apply_splits_to_sorting(
    280             curated_sorting_or_analyzer,
    281             split_units,
   (...)
    284             return_extra=True,
    285         )
    286     else:
--> 287         curated_sorting_or_analyzer, _ = curated_sorting_or_analyzer.split_units(
    288             split_units,
    289             new_id_strategy=new_id_strategy,
    290             return_new_unit_ids=True,
    291             new_unit_ids=split_new_unit_ids,
    292             format="memory",
    293             verbose=verbose,
    294         )
    296 return curated_sorting_or_analyzer

File ~/arbeit/PyCode/spikeinterface/spikeinterface-git/src/spikeinterface/core/sortinganalyzer.py:1426, in SortingAnalyzer.split_units(self, split_units, new_unit_ids, new_id_strategy, return_new_unit_ids, format, folder, verbose, **job_kwargs)
   1423 new_unit_ids = generate_unit_ids_for_split(self.unit_ids, split_units, new_unit_ids, new_id_strategy)
   1424 all_unit_ids = _get_ids_after_splitting(self.unit_ids, split_units, new_unit_ids=new_unit_ids)
-> 1426 new_analyzer = self._save_or_select_or_merge_or_split(
   1427     format=format,
   1428     folder=folder,
   1429     split_units=split_units,
   1430     unit_ids=all_unit_ids,
   1431     verbose=verbose,
   1432     split_new_unit_ids=new_unit_ids,
   1433     **job_kwargs,
   1434 )
   1435 if return_new_unit_ids:
   1436     return new_analyzer, new_unit_ids

File ~/arbeit/PyCode/spikeinterface/spikeinterface-git/src/spikeinterface/core/sortinganalyzer.py:1154, in SortingAnalyzer._save_or_select_or_merge_or_split(self, format, folder, unit_ids, merge_unit_groups, censor_ms, merging_mode, sparsity_overlap, merge_new_unit_ids, split_units, splitting_mode, split_new_unit_ids, backend_options, verbose, **job_kwargs)
   1151 else:
   1152     # split
   1153     if splitting_mode == "soft":
-> 1154         new_sorting_analyzer.extensions[extension_name] = extension.split(
   1155             new_sorting_analyzer, split_units=split_units, new_unit_ids=split_new_unit_ids, verbose=verbose
   1156         )
   1157     elif splitting_mode == "hard":
   1158         recompute_dict[extension_name] = extension.params

File ~/arbeit/PyCode/spikeinterface/spikeinterface-git/src/spikeinterface/core/sortinganalyzer.py:2628, in AnalyzerExtension.split(self, new_sorting_analyzer, split_units, new_unit_ids, verbose, **job_kwargs)
   2626 new_extension = self.__class__(new_sorting_analyzer)
   2627 new_extension.params = self.params.copy()
-> 2628 new_extension.data = self._split_extension_data(
   2629     split_units, new_unit_ids, new_sorting_analyzer, verbose=verbose, **job_kwargs
   2630 )
   2631 new_extension.run_info = copy(self.run_info)
   2632 new_extension.save()

File ~/arbeit/PyCode/spikeinterface/spikeinterface-git/src/spikeinterface/core/analyzer_extension_core.py:607, in ComputeTemplates._split_extension_data(self, split_units, new_unit_ids, new_sorting_analyzer, verbose, **job_kwargs)
    605 new_indices = np.array([new_analyzer_unit_ids.index(unit_id) for unit_id in unsplit_unit_ids])
    606 old_indices = self.sorting_analyzer.sorting.ids_to_indices(unsplit_unit_ids)
--> 607 new_array[new_indices, ...] = arr[old_indices, ...]
    609 for split_unit_id, new_splits in zip(split_units, new_unit_ids):
    610     if new_sorting_analyzer.has_extension("waveforms"):

IndexError: arrays used as indices must be of integer (or boolean) type

From what I gathered, in line 604 in analyzer_extension_core.py the unsplit_unit_ids is empty if there are no units left which were not split. Tried a lazy fix

if unsplit_unit_ids:
            new_indices = np.array([new_analyzer_unit_ids.index(unit_id) for unit_id in unsplit_unit_ids])
            old_indices = self.sorting_analyzer.sorting.ids_to_indices(unsplit_unit_ids)
            new_array[new_indices, ...] = arr[old_indices, ...]

but that results in a whole bunch of problems downstream. The possible workaround is to simply make sure that there is at least one unit left un-deleted and un-split in a given curation session, and then split or delete it in a new curation session.

spikeinterface v. 0.104.1
spikeinterface-gui v. 0.13.1