SpikeInterface
Start porting some lussac concept in curation modules.
This looks good!
If you look at the performance, there might be a few to a lot of false positives (depending on the recording), which can usually be sorted out by looking at the template waveform.
For the reshifting, I don't think it's necessary to compare the correlograms (usually the bin size is large enough), but it is very important if you merge the 2 spike trains together! I think just recentering the units on their own and then calling this function will be enough.
I think a low-pass filter at 300 Hz is too agressive (I have to change it in the default params of Lussac actually). If you have units with a refractory_period of 1ms, then the frequencies around 500-1000 Hz are very important. Maybe set it at something like 800 Hz by default?
@alejoe91 : this is ready to merge on my side. We will be able to improve later.
@DradeAW : ok for you ?
There's still the test_automerge in the root folder that needs to be deleted (I erroneously pushed it),
Other than that I think it's good for now, I'll work on some optimization in the future.
There's still the
test_automergein the root folder that needs to be deleted (I erroneously pushed it), Other than that I think it's good for now, I'll work on some optimization in the future.
I do not see it anymore.
The fail is due to a bug I fixed in #1061
The method 'random' for remove_duplicated_spikes sometimes isn't doing what it should do.