spikeinterface: `mountainsort5` failing due to change in `from_times_labels`
MRE
import spikeinterface.full as si
import probeinterface as pi
import numpy as np
recording, _ = si.generate_ground_truth_recording(durations=[600], sampling_frequency=30_000, num_channels=128)
# this code is for the Cambridge neurotech option to glue to probes together which naturally fits well with a probe group :)
probe = pi.get_probe("cambridgeneurotech", "ASSY-77-H5")
probe1 = probe.copy()
probe1.move([650,0])
wiring = np.arange(0,128)
multishank = pi.ProbeGroup()
multishank.add_probe(probe)
multishank.add_probe(probe1)
multishank.set_global_device_channel_indices(wiring)
recording = recording.set_probegroup(multishank)
# adding this because my normal work flow removes some channels based on bad impedance values.
# I don't know if this matters yet.
recording = recording.remove_channels(['10', '35', '67'])
ms_sorting = si.run_sorter_by_property(sorter_name="mountainsort5", recording=recording, grouping_property="group", folder='test')
Error
This fails with the following error
extract_snippets
snippets[j] = traces[t1:t2]
~~~~~~~~^^^
ValueError: could not broadcast input array from shape (0,64) into shape (40,64)
But this happens no matter what data I’m working with (both simulated or real data). On version 0.102.1 this error does not happen.
Actual Problem
Not a clue yet. Unfortunately the error happens in sorting so I’m assuming something goes wrong either in channel slicing or in preprocessing. Mountainsort5 internally runs our bandpass_filter and our whiten so either of those functions since the update to preprocessing could be the problem. I need to run more tests. But I’m pretty busy today, so if someone else wants to pick this up too that would be great.
Environment
spikeinterface: main branch
neo: main branch
numpy 1.26.0
scipy 1.11.3
python 3.11.6
EDIT typed out my script too fast… edited to fix 😃
About this issue
- Original URL
- State: closed
- Created 2 years ago
- Comments: 15 (15 by maintainers)
Update. Shutting off filter and whiten in mountainsort5 did not fix the issue. Removing the
remove_channelsdid not fix it.