Self matching don't improve above 15%

[taningh86]

Hi,
I am wondering why my self-matching (comparison between the the two splits) never improves above 15%. As a result I get very high false negatives, basically 85%. This suggests either something is seriously wrong in the waveform extraction process (spikeinterface demo notebook) or my data is noisy. The latter is not true because my data yield are generally standard and robust, and I have tracked more than 40% of neurons manually using Phy across months (that is also using very stringent protocol). The KS outputs itself labels about ~70-80% as 'good', which I don't take it for granted and always verify them manually. So, I am not sure where the issue is.

Also, I have found quite a few errors in your spikeinterface demo notebooks that leads to errors. If you are curious I can give them out to you so people who are not very adept in python are not able to troubleshoot or don't get stuck. also, I am happy to share my notebook for you to see where things could be different than yours and diagnose why self-matching itself is poor.

Thanks
Jimmy

[EnnyvanBeest]

Hi Jimmy,

apologies, I haven't checked this page for a while as we were busy releasing DeepUnitMatch. We have made quite some changes and hopefully clarifications as part of this release. Could you please let me know if things are still unclear or whether all your previous concerns are now solved?

Thanks,
Enny