Alessio Buccino

@WeissShahaf fixed here #3139, but with a slight different design: You can optionally pass a list of channels to compute the median on with the `ref_channel_ids` existing argument

Hi @DaohanZhang Can you plot a spectrum of the traces? If the noise is localized, you can play with the q factor of the notch filter to broaden/narrow the notch...

It would be helpful to look at the spectra though (couldn't find them on slack)

Thanks, can you also plot the power spectrum density? It makes it easier to check for very localized noise sources. From the spectrograms, they seem quite distributed so a notch...

@DaohanZhang Can you try a notch filter where you have the peak (~5000Hz) and see if it improves the situation?

Hi @DavidLucha Are you applying some preprocessing on the SpikeInterface side? You would need at least a high-pass/band-pass filter before creating the analyzer, otherwise everything will be off due to...

Yes, you still need at least a highpass filter because different channels usually have different offsets :) Give it a try, and you'll see!

Highpass + CMR are quite standard and highly recommended to get some denoised signals! I think this is good enough in most of the cases

Dprime and nn metrics depend a lot on the neighbor units. If the brain area is very "populated", you'll get much smaller values. Can I recommend trying out the UnitRefine...

> Perfect, fine by me. But then, assuming you have too many spikes to be in memory, how do you see the rest of the algorithm? Can we still load...