spikeinterface
spikeinterface copied to clipboard
SpikeInterface
Correlogram gaussian
Implementing the code for gaussian-filtered correlograms computed in numba.
Implementation is done.
However:
1- No tests have been made yet (I will do them)
2- compute_gaussian_correlograms is pretty slow right now. For a sorting with ~200 units (ks2 output) of a 45 min recording generated with MEArec, it takes around 8m30. I'll see if I can optimize some parts of the code.
With some changes to the default parameters and some code re-writing, I managed to go down from 8m30 to 40s. Still takes a bit of time, but it's much better.
The next step for optimization would be to look at making a convolution in the fourier domain, but I don't have time to look at this. I'll make the tests, and this PR will be good to go on my end.
The more I think about it, the more I think we don't need a special function for gaussian correlograms.
People can just call the regular function with a bin_size of 1 sample and to the gaussian filtering themselves.