Skip to main content

How to compute features from SpikeInterface Zarr stores

This guide computes features from a recording saved in the SpikeInterface-native Zarr layout — the format behind, for example, the AIND / open-ephys WavPack-compressed ecephys_compressed/*.zarr stores. Local paths and remote object stores (S3, GCS) work the same way.

Once the calculator is built, features are computed as for SpikeGLX files, with compute_denoised_features. Only the calculator changes.

ImportantNot for NWB files with a Zarr backend

SpikeInterfaceZarrFeatureCalculator reads a ZarrRecordingExtractor store — the layout spikeinterface.core.read_zarr produces, with traces_seg0 and friends inside.

An NWB file written with a Zarr backend (.nwb.zarr) is a different format despite also being Zarr on disk. Read those with NwbFeatureCalculator and backend="zarr". Using the wrong calculator produces an opaque reader error.

Prerequisites

SpikeInterface ships in the [full] extra:

pip install -e ".[full]"

Two further packages depend on where and how your store is written, and are not installed by either extra:

You need When
s3fs Reading a store from S3
wavpack_numcodecs Decompressing AIND WavPack stores 
pip install s3fs wavpack_numcodecs

wavpack_numcodecs must be imported before reading, so it registers its codec with numcodecs:

import wavpack_numcodecs  # noqa: F401  — registers the WavPack codec

A public store, streamed from S3

storage_options is forwarded straight to read_zarr. Use {"anon": True} for unauthenticated reads of a public bucket:

import wavpack_numcodecs  # noqa: F401  — registers the WavPack codec

from ephysatlas.feature_calculators import SpikeInterfaceZarrFeatureCalculator

S3_LF = (
    "s3://aind-open-data/ecephys_830794_2026-01-26_12-02-05/ecephys/"
    "ecephys_compressed/experiment1_Record Node 101#Neuropix-PXI-100.ProbeA-LFP.zarr"
)

calc = SpikeInterfaceZarrFeatureCalculator(
    lf_zarr=S3_LF,
    storage_options={"anon": True},
    name="aind_830794_ProbeA",
)

df = calc.compute_denoised_features(n_snippets=5, duration_ap=0.0, duration_lf=5.0)
print(df[["rms_lf", "psd_delta"]].head())

Local stores

Local stores need no storage_options. Point the calculator at both bands to get the full feature set:

calc = SpikeInterfaceZarrFeatureCalculator(
    ap_zarr="/data/ecephys_compressed/experiment1_...ProbeD-AP.zarr",
    lf_zarr="/data/ecephys_compressed/experiment1_...ProbeD-LFP.zarr",
    name="probeD",
)

df = calc.compute_denoised_features(n_snippets=5, duration_ap=5.0, duration_lf=5.0)
print(df.index.names)   # ['pid', 'channel']

AP and LF live in separate stores in this format — unlike NWB, where one file usually holds both bands. Note the -AP / -LFP suffixes above.

At least one of ap_zarr / lf_zarr is required; supplying neither raises ValueError. Stores are opened lazily on first access, so constructing the calculator is cheap.

To compute a single snippet, choose the windows yourself, keep the snippet files, or work with a recording shorter than about 10 minutes, see tune the computation.

Naming the output

name defaults to the stem of the first available store, which for these paths is long and awkward. Set it explicitly — it becomes the first index level of the feature table and the probe-level output directory:

calc = SpikeInterfaceZarrFeatureCalculator(lf_zarr=S3_LF, name="aind_830794_ProbeA")

Probe version

neuropixel_version (default 1) is forwarded to destriping. Set 2 for NP2 probes:

calc = SpikeInterfaceZarrFeatureCalculator(
    lf_zarr=..., name=..., neuropixel_version=2
)

Troubleshooting

ValueError: At least one of ap_zarr or lf_zarr must be provided — neither store was given.

A numcodecs error about an unknown codec wavpack — you did not import wavpack_numcodecs before reading. Import it at the top of your script.

A filesystem error on an s3:// path — install s3fs. If the bucket is public, also pass storage_options={"anon": True}; without it, read_zarr tries to sign the request and fails on missing credentials.

A reader error mentioning missing groups or attributes — you may have an NWB-Zarr file rather than an SI-native store. See the callout at the top of this page.

NotImplementedError: ... recording spans 4 shanks/groups ... — the store’s channel groups cover more than one shank, and feature computation is single-shank. Give each calculator one shank’s channels; see one shank per calculator.

Next steps