search

csn-le / wave_clus

A fast and unsupervised algorithm for spike detection and sorting using wavelets and super-paramagnetic clustering
121 stars 65 forks source link

Wave Clus Overclassification #202

Closed slassers closed 2 years ago

slassers commented 2 years ago

Hi Fernando,

Comparison of delay times between two adjacent electrodes produced a constant delay time suggesting a single axon arising from three clusters on each electrode. How can we get wave clus to automatically combine these clusters? Lassers NMI Results Allowing Multiple Matches Per Cluster 201025gb

We have tried modulating a variety of parameters from temperature, overlap, threshold border parameter, max and min standard deviation, and minimum cluster size.

If you have any ideas for automatic merging it would be greatly appreciated! -Sam

ferchaure commented 2 years ago

Hi Sam, you have a lot of noise in the A7 channel, I would increase stdmin

Merging could be difficult, The waveforms look different, you have some type of noise... maybe a 25Hz noise, check for peaks in the spectrum.

An alternative is reducing w_pre to 14 and w_pos to 18 just to have smaller waveforms without that artefacts