search

flatironinstitute / ironclust

Spike sorting software being developed at Flatiron Institute, based on JRCLUST (Janelia Rocket Cluster)
Apache License 2.0
28 stars 7 forks source link

Question about single channel detection #54

Open ferchaure opened 4 years ago

ferchaure commented 4 years ago

Hi, I have been testing Ironclus using spikeinterface and toy examples with all the spikes of the neurons at least one sample over 5*MAD (I disable all the filters just to check the detection). But It looks like the algorithm can't detect some classes, usually the ones that are closer to the threshold. The parameters I changed are: 

'detect_sign' : 1,
'whiten': False,
'filter_type':'none',
'filter_detect_type':'none',
'clip_pre':0.8,
'clip_post':1.8,
'filter': False,
'detect_threshold': 5,
'common_ref_type':'none',

Do you know why this could be happening? Is some criteria applied to remove detections before assigning the spikes to a class?

ferchaure commented 4 years ago

I made more test and the issue is related to the detect_sign parameter. Ironclus is using a negative detection regardless of the setting.