search

CINPLA / CA2_MEC

GNU General Public License v3.0
0 stars 1 forks source link

Identify_neurons: not same result on educloud #3

Open sarahthon opened 1 month ago

sarahthon commented 1 month ago

Hi, after spikesorting + curating in phy + saving to exdir, identify_neurons in the old repository (which runs locally) is able to find 12 cells across all 5 recordings, as you can see on notebooks/00-identify-neurons.ipynb at vemundss branch in the old repository (https://github.com/sarahthon/CA2_MEC/blob/vemundss/notebooks/00-identify-neurons.ipynb).

When spikesorting the same five recordings (001-181220-2, 001-181220-3, 001-181220-4, 001-181220-5, 001-181220-6) on educloud (project path: Sarah/projects/hpc-social-nwb), identify_neurons in this repository only finds 6.

lepmik commented 1 month ago

Permission denied: '/projects/ec109/Sarah/projects/hpc-social-nwb/actions/identify-neurons/data/sessions.csv'

alejoe91 commented 1 month ago

@sarahthon one suggestion from Mikkel is to compute templates using all waveforms.

Is this project a conversion from an old exdir project?

sarahthon commented 1 month ago

@alejoe91 yes, these are converted.

alejoe91 commented 1 month ago

I'll push a fix there, but you might need to reconvert

alejoe91 commented 1 month ago

@sarahthon why did you close the issue? :)

sarahthon commented 1 month ago

Ops, sorry!

alejoe91 commented 1 month ago

Ok, in this last commit all waveforms are extracted, so it should give a better estimate of the templates used for unit tracking

I would wait for @lepmik to deploy it on EduCloud. Is that ok?

@sarahthon just for clarification: you did not re-spikesort the recordings, right? Otherwise the mismatch could be very well due to different spike sorting outcomes

sarahthon commented 1 month ago

@alejoe91 I only converted + «save to NWB units» for these recordings.

I do plan on spikesorting new actions and running identify_neurons on educloud later.

sarahthon commented 3 weeks ago

We have tested identify_neurons with the newst updates. We still don't get the same results, but I now get 7 out of the 12 cells that match locally.

lepmik commented 3 weeks ago

Then I'm out of ideas and out of time, @alejoe91 could you look into this?

alejoe91 commented 3 weeks ago

@sarahthon can you try to check the old exdir templates and the new ones one by one? At least for a session. If they are exactly* the same then it must be some package versioning difference, but these would be very hard to find..