search

cortex-lab / KiloSort

GPU code for spike sorting
GNU General Public License v2.0
179 stars 100 forks source link

Still missing spikes #180

Open MarReus opened 5 years ago

MarReus commented 5 years ago

Hi guys! I'm facing the same problem! (#84, #117, #160)

untitled1

untitled2

I checked and It's not a visualization problem (you can also see it in the screen shots) and I also tried changing all those parameters: I used a 16 channels silicon prove, and my recordings are around 100 sec long, in awake, behaving animals, so sometimes are a little noisy. In the begging I used 32 clusters, but then I changed to 64 to try if it was better, but I still have a lot of spikes out of the clustering. I also use ops.initialize = 'fromData'; ops.spkTh = -4, and played with ops.Th = [4 8 8]; and ops.lam = [10 30 30] but still missing a lot.

I also realized that the remaining recording after subtracting the template (the non colored one) is no as flat as it should be, so I'm assuming something is wrong with the clustering.

Does somebody have any other advise for me? I would really appreciate! Thanks in advance!

marius10p commented 5 years ago

Sorry, I don't think we ever figured out where this behavior was coming from. Before the end of the month I will release Kilosort2, and if this problem persists, I can take a look at your data.

chris-angeloni commented 5 years ago

@marius10p great to hear that KS2 will be released soon! Looking forward to it.

@MarReus Have you tried putting ops.Th even lower? Say [2 6 6]? Not sure if it would help though, because it's missing spikes that have the same amplitude as marked ones.

Dakshitha-BA commented 5 years ago

@MarReus Did you figure out the problem? I am facing the same issue.

susuchen66 commented 5 years ago

Hi we are facing the same issue with KS2 missingspk3 missingspk4

kingsEffy commented 9 months ago

I'm facing the same using kilosort3...