Open charlesdgburns opened 9 hours ago
@Sam-Dodgson I think there may be a bug here: it seems it still loads in only the good units despite bool=false? Can you double check please?
To OP: It will most likely not work well including all units including noise units, as the data within day will be too unreliable to find a proper threshold for reliable matches. So even when the waveform bug is fixed, you may find that same error message. So using UM this way is not recommended.
Thanks for the quick response!
Regarding the inclusion of noise units - I also realised the point you've made after hacking the kilosort labels to all be 'good'.
Is there any way for unitmatch to label these as noise itself? Or do you generally find that the kilosort 'good' label always excludes such noisy units?
Using the Kilosort label is a good start, at least to throw out the noise. However Kilosort only decides on 'good' versus 'mua' based on the ISI distribution. We personally use bombcell but there are other quality metrics softwares as well. Manual curation with Phy is another option, but with a lot of data may become too labour intensive.
Was curious to see how unit match would run across all units. I'm running code as is set up in the jupyter notebook.
and I run into a slightly obscure error at the stage of
error msg:
So far I've identified a problem with the output of
uitl.load_good_waveforms
whengood_units_only = False
.It's clear for average waveforms after running
ov.extract_parameters()
, as I only get two different waveforms across all clusters (whengood_units_only = False
):but when good_units_only = True, I get a different average waveform for each cluster: