SpikeInterface 0.104.0 release notes

Mar 23rd 2026

Main changes:

• Added support for Bombcell [Fabre], for unit curation, including new and updated template metrics (#4306)

• Added LUPIN [Garcia2024] as an internal sorter (#4192)

• Added support for SLAy [Koukuntla], for automated unit merges (#4190)

• Added valid_unit_periods postprocessing extension, based on NeuroPyxles [npyx] (#4299, #4302)

• Added AutoCorrelograms postprocessing extension (#4307)

• Added support for double-sided probes (#4272)

• Added SequentialCuration, allowing for chains of curations to be applied to analyzers (#4298)

• Added support for FigPack, to eventually superseed SortingView (#4364)

• Major refactor and debugging of the Metrics postprocessing module (#4183, #4362, #4355, #4353, #4342, #4292) (see: Metrics module)

• Significant speedups in Metrics (#4308, #4363) and SortingComponents (#4402, #4341) modules, and in correlograms (#4305) and template similarity (#4343) computations.

• API and breaking changes:

  • Specific metrics functions now imported from different sub module (see: Metrics module)

  • snr quality metric calculation now uses median rather than mean (see: Quality Metrics module)

  • Metric name “peak_to_valley” has been renamed to “peak_to_trough_duration” for clarity. This name change will be applied automatically on load of older analzyers. (see: Template Metrics)

  • The function to do UnitRefine curation has been renamed from auto_label_units to unitrefine_label_units (see unitrefine_label_units())

  • Remove support for Python 3.9

core:

• Fix closest_channels sparsity estimation when n_units > n_channels (#4443)

• Copy the recording from original sorting during NumpySorting.from_sorting() (#4417)

• Don’t raise when extension is unknown (#4406)

• Make NumpySorting.from_unit_dict docs and code consistent (#4382)

• Improve extension dependency management in SortingAnalyzer (#4321)

• Implement select_sorting_periods in core (#4316)

• Implement the cached_lexsorted_spike_vector. (#4315)

• Move structured dtypes to base (#4314)

• Implement select_sorting_periods in metrics (#4302)

• Fix get_probe set_probe annotations propagation (#4300)

• using spike_vector on core/check_sortings_equal test (#4282)

• Random spike selection new methods (#4276)

• Implement the by_side grouping with probes (#4272)

• Fix gaussian filter with time vector (#4268)

• Delete some unused references to channel_slice (deprecated in v0.103.0) (#4246)

• Fix probable sharedmem (#4239)

• Implement Add/SubtractRecording classes and +- operators (#4238)

• Warn about non-mergeable units instead of raising an error (#4219)

• Implementing sparsity for ComputeTemplates (#4212)

extractors:

• Add XClustSortingExtractor (#4433)

• Add gain and offset for read_kilosort_as_analyzer (#4428)

• Fix openephys tests (#4426)

• Fix Open Ephys probe channel order (#4396)

• Improve NWB Extractor __del__ method to avoid import errors (#4369)

• Fix IBL test with probe representation (#4335)

• Remove check of len(inter_sample_shifts) matching num_readout_channels (#4317)

• Update to SiNAPS extractor (#4263)

• Support for using event names as IDs in PL2 files (#4049)

• Get unit locations from HerdingspikesSortingExtractor (#3525)

preprocessing:

• Slice epsilons with dead channels to match gains during agc. (#4430)

• Support time_vector in resample() (#4429)

• Implementing KNN referencing and a backup referencing mechanism for common_reference (#4412)

• Use a truncated median filter near edges when detecting bad channels (#4391)

• Make anyone happy about warnings and margin in filtering. (#4384)

• Fix bug spatial filter #4175 (#4286)

• Fix gaussian filter with time vector (#4268)

• Fixes Issue4180 (#4247)

• Fix artifact removal padding logic (#4245)

• Auto-estimate margins based on freq_min/f0+Q for filters (#4227)

• Fix silence period and job_kwargs (#4071)

• Remove artefacts based of an envelope (#3715)

sorters:

• Add the option for clustering methods to return temporal shifts for peaks while clustering (#4401)

• Add plotting functions for drift and diagnostics to KS4 (#4387)

• lupin and tdc2 : improvements (#4349)

• Improve multithreading and multi processing : lock + nogil + mp_context (#4333)

• Fix typo in lupin/tdc2 (#4284)

• Make “simple” sorters params simpler. (#4281)

• Improvements for internal sorters (#4279)

• kilosort4 may fail due to the undesirable instantiation of logger (#4277)

• Flatten params for tridesclous2 sorter (#4254)

• Add isosplit in simple sorter (#4249)

• Final cleaning of mixtures for lupin/tdc/sc (#4244)

• Cosmetic logging messages for SC2 (#4242)

• A new sorter : Lupin sorter (#4192)

postprocessing:

• Set amplitude scalings equal zero for zero template units (#4455)

• Add gain and offset for read_kilosort_as_analyzer (#4428)

• Don’t raise when extension is unknown (#4406)

• Fix valid period amplitudes (#4361)

• Patch lag similarities (#4345)

• Speed up template similarity computing using numba (#4343)

• Change ComputeSpikeLocations to use only peak_sign (#4330)

• Ensure compute unit locations respects order of unit ids (#4329)

• Auto correlograms (#4307)

• Option to turn on parallel computation for CCG in numba (#4305)

• Valid unit periods extension (#4299)

• Add BaseSpikeVectorExtension (#4189)

metrics:

• Set num_histogram_bins in amplitude_cutoff (back) to 100 (#4415)

• Fix metrics test for numba correlograms fast_mode (#4377)

• Remove multiprocessing from NearestNeigbhor metric (#4363)

• Add peak_sign to sd_ratio (#4362)

• Improve error message for deprecated metric names (#4358)

• Correct formula for mad (and std) in drift quality metrics (#4355)

• Compute missing spikes on the right tail of amplitude distributions (#4353)

• Improve stability of velocity fits in template metrics (#4342)

• Timing the computations of quality metrics and some optimizations (#4308)

• Bombcell integration and template metrics refactoring (#4306)

• Fix rp violations bug (#4292)

• Refactor metrics into its own module (#4183)

curation:

• Improve robustness of model classifier tests (#4447)

• Rename CurationModel to Curation (#4421)

• Change ‘min’/’max’ to ‘greater’/’less’ when defining thresholds for threshold_metrics_label_units (#4416)

• Send peak_sign to get_template_extremum_channel_peak_shift in remove_redundant_units (#4408)

• Use sorting._recording (if available) in remove_excess_spikes (#4407)

• Add threshold_metrics curation (#4365)

• Ensure default supported_versions is a tuple, as in type (#4348)

• Refactor API for auto_label_units (#4338)

• Bombcell integration and template metrics refactoring (#4306)

• Sequential curation (#4298)

• Fix failing tests due to scikit-learn versions (#4253)

• Warn about non-mergeable units instead of raising an error (#4219)

• Add SLAy auto-merge preset (#4190)

widgets:

• Enlarge channel textbox in plot_traces (#4400)

• Send a float (rather than 1-element array) to BoundedFloatText (#4399)

• Figpack integration (replacement of sortingview) (#4364)

• Remove layer selector from plot_traces ipywidgets backend if single layer (#4325)

• Fix edges of 2d-histogram in UnitWaveformDensityMapWidget (#4283)

• Protect unit summary and unit locations plots against NaNs (#4128)

generators:

• More possibilities for gt generation (#4289)

• Update generate.py (#4163)

sortingcomponents:

• Optimizations: faster waveforms/templates and seeding iterative_isosplit properly (#4402)

• Add the option for clustering methods to return temporal shifts for peaks while clustering (#4401)

• Make peak detection (locally_exclussive, matched_filtering) faster and more accurate. (#4341)

• Needed by Olivier to create an analyzer with precomputed data (#4303)

• Final cleaning of mixtures for lupin/tdc/sc (#4244)

• Cosmetic logging messages for SC2 (#4242)

• Update nearest peeler (#4223)

documentation:

• Add missing API doc entries for save*, dump*, remove_channels, and scale_to_uV (#4451)

• Extend unsigned to signed docs (#4448)

• Fix link to LFP docs in error message (#4424)

• Fix some typos in documentation (#4419)

• Add a why integrate your software with spikeinterface section (#4393)

• Add compute_valid_unit_periods to api and add credit in postprocessing module (#4390)

• Add metric_descriptions and BaseMetricExtension.get_metric_colum_descriptions (#4278)

• Doc sorters (#4275)

• Clean auto-release script: use metrics + only print changed modules (#4265)

continuous integration:

• Update setuptools version for legacy kilosort (#4386)

• Get rid of pytest-dependency and use fixtures (#4381)

• Free up space in codecov CI (#4252)

packaging:

• Update lower bound python support to 3.10 (#4383)

testing:

• Further Improve robustness of model classifier tests (#4450)

• Remove n_jobs=-1 from curation tests (#4449)

• Improve robustness of model classifier tests (#4447)

• Fix openephys tests (#4426)

• Rename neurodsp to ibldsp to ensure IBL tests run (#4392)

• Update setuptools version for legacy kilosort (#4386)

• Get rid of pytest-dependency and use fixtures (#4381)

• Test IBL extractors tests failing for PI update (#4285)

• using spike_vector on core/check_sortings_equal test (#4282)

• fix tdc2 failing tests (#4271)

• Fix failing tests due to scikit-learn versions (#4253)

deprecations:

• Remove deprecations for 0.104.0 (#4425)

• Make anyone happy about warnings and margin in filtering. (#4384)

• Improve error message for deprecated metric names (#4358)

• Binary recording bacward-compatibility (#4357)

Contributors:

• @CodyCBakerPhD

• @JoeZiminski

• @Julie-Fabre

• @MGAMZ

• @OleBialas

• @ShijiMi-Soup

• @ablot

• @alejoe91

• @b-grimaud

• @chrishalcrow

• @ecobost

• @esutlie

• @grahamfindlay

• @h-mayorquin

• @m-beau

• @mhhennig

• @oliche

• @pas-calc

• @raffaele222

• @samuelgarcia

• @tayheau

• @ueeseer

• @yger

• @zzhmark