`SortGroup.set_group_by_shank` crashes on non-numeric `electrode_group_name` (hard-coded `int()` sort)

[pauladkisson]

Describe the bug

SortGroup.set_group_by_shank(nwb_file_name=...) raises ValueError: invalid literal for int() with base 10: ... for any session whose electrodes have a non-numeric electrode_group_name.

The helper get_group_by_shank (in spikesorting/utils.py) sorts the electrode-group names with e_groups.sort(key=int), which assumes every group name is a numeric string (e.g. "0", "1"). NWB files that name electrode groups descriptively — ours use "probe1_shank1" — fail the int() cast.

This blocks the standard sort-group setup step entirely; the only workaround is to bypass set_group_by_shank and insert SortGroup / SortGroup.SortGroupElectrode rows manually.

To Reproduce

This script is self-contained: it builds a tiny NWB file with an ndx-franklab-novela Probe whose electrode group is named "probe1_shank1", inserts it into Spyglass, and then calls set_group_by_shank. (Set dj_local_conf_path to your own config.)

"""Minimal reproduction: SortGroup.set_group_by_shank crashes on a non-numeric
electrode_group_name."""
from pathlib import Path
import numpy as np
import datajoint as dj

# Point this at your own DataJoint config (or remove if already configured elsewhere).
dj_local_conf_path = "dj_local_conf.json"
dj.config.load(dj_local_conf_path)
dj.config["enable_python_native_blobs"] = True

from pynwb import NWBHDF5IO
from pynwb.testing.mock.file import mock_NWBFile, mock_Subject
from ndx_franklab_novela import Probe, Shank, ShanksElectrode, NwbElectrodeGroup

import spyglass.common as sgc
import spyglass.data_import as sgi
import spyglass.spikesorting.v1 as sgs
from spyglass.settings import raw_dir
from spyglass.utils.nwb_helper_fn import get_nwb_copy_filename


def build_nwbfile():
    nwbfile = mock_NWBFile(
        identifier="set_group_by_shank_bug",
        session_description="Mock NWB demonstrating set_group_by_shank int() bug",
    )
    mock_Subject(nwbfile=nwbfile, subject_id="set_group_by_shank_subject")

    n_channels = 4
    shanks_electrodes = [
        ShanksElectrode(name=str(i), rel_x=0.0, rel_y=0.0, rel_z=float(i))
        for i in range(n_channels)
    ]
    # Shank name must match the electrodes' probe_shank value below (Spyglass keys
    # Probe.Electrode on (probe_id, probe_shank, probe_electrode)).
    shank = Shank(name="1", shanks_electrodes=shanks_electrodes)
    # Novel probe_type / name so the example does not collide with any existing probe.
    probe = Probe(
        name="test_probe_nonnumeric_group",
        id=1,
        probe_type="test_probe_nonnumeric_group",
        probe_description="1x4 test probe",
        contact_side_numbering=True,
        contact_size=10.0,
        units="um",
        shanks=[shank],
    )
    nwbfile.add_device(probe)

    # NON-numeric electrode group name -- this is what trips set_group_by_shank.
    # Use the franklab NwbElectrodeGroup, as real franklab/novela files do.
    electrode_group = NwbElectrodeGroup(
        name="probe1_shank1",
        description="test shank",
        location="hippocampus",
        device=probe,
        targeted_location="hippocampus",
        targeted_x=0.0,
        targeted_y=0.0,
        targeted_z=0.0,
        units="um",
    )
    nwbfile.add_electrode_group(electrode_group)

    # Custom columns Spyglass expects from the franklab/novela electrode format.
    for col in ["probe_shank", "probe_electrode", "bad_channel", "ref_elect_id"]:
        nwbfile.add_electrode_column(name=col, description=col)
    for i in range(n_channels):
        nwbfile.add_electrode(
            id=i, group=electrode_group, location="hippocampus",
            probe_shank=1, probe_electrode=i, bad_channel=False, ref_elect_id=-1,
        )
    return nwbfile


def main():
    nwbfile_path = Path(raw_dir) / "set_group_by_shank_bug.nwb"
    if nwbfile_path.exists():
        nwbfile_path.unlink()
    with NWBHDF5IO(nwbfile_path, "w") as io:
        io.write(build_nwbfile())

    nwb_copy_file_name = get_nwb_copy_filename(nwbfile_path.name)
    (sgc.Nwbfile & {"nwb_file_name": nwb_copy_file_name}).delete(safemode=False)
    sgi.insert_sessions(str(nwbfile_path.name), rollback_on_fail=True, raise_err=True)

    print(
        "electrode_group_name values:",
        np.unique(
            (sgc.Electrode & {"nwb_file_name": nwb_copy_file_name}).fetch(
                "electrode_group_name"
            )
        ),
    )

    sgs.SortGroup.set_group_by_shank(nwb_file_name=nwb_copy_file_name)
    # --> ValueError: invalid literal for int() with base 10: 'probe1_shank1'


if __name__ == "__main__":
    main()

Output (the insert succeeds, then set_group_by_shank fails):

electrode_group_name values: ['probe1_shank1']

Error Stack

Traceback (most recent call last):
  File "repro_set_group_by_shank.py", line 100, in <module>
    main()
  File "repro_set_group_by_shank.py", line 95, in main
    sgs.SortGroup.set_group_by_shank(nwb_file_name=nwb_copy_file_name)
  File "/.../spyglass/src/spyglass/spikesorting/v1/recording.py", line 88, in set_group_by_shank
    sg_keys, sge_keys = get_group_by_shank(
  File "/.../spyglass/src/spyglass/spikesorting/utils.py", line 50, in get_group_by_shank
    e_groups.sort(key=int)  # sort electrode groups numerically
ValueError: invalid literal for int() with base 10: 'probe1_shank1'

Suggested fix

get_group_by_shank only needs a stable ordering of the group names; the numeric assumption is incidental. Replace the hard-coded int key with something that tolerates non-numeric names, e.g.:

try:
    e_groups.sort(key=int)
except ValueError:
    e_groups.sort()  # fall back to lexicographic order for non-numeric group names

(or a natural-sort key if numeric-aware ordering of names like "shank2" vs "shank10" matters).

Additional context

• Spyglass 0.5.5a2.dev2+g407cd6624 (commit b6d59a0, 2026-06-05), DataJoint 0.14.9, Python 3.10.
• Source: spikesorting/utils.py:50, in get_group_by_shank, called by SortGroup.set_group_by_shank (spikesorting/v1/recording.py:51).
• Suggested labels: bug, spike sorting.