Open subhacom opened 4 years ago
In openelextrophy a very old code I did long time ago I tested this wavelet feature project by copy/paste from wave_clust. (more or less 10 lines of code) I did not see any improvement. So I did not implement this in tridesclous.
A strength of tridesclous is that you can add very easily new methods for each steps (cluster, feature, peeler, that a comnibaison of methods for steps can be tested without pain).
Have a look to tridesclous/decomposition.py, you will see that that it is super easy to add new method like this wavelet tranform compression. I can help you if you want. Or if you send me an example in python or matlab I can do it my self also. We basically have implemnt a class tha follow the fit/predict workflow of sklearn.
In many cases where human eye can clearly see difference in spike shapes, after PCA they all get projected into a single blob that cannot be separated by clustering. Wavelet transform has been shown to perform better in such cases, but very few spike sorters allow this (e.g. wave_clus and combinato).
It will be great to have the option of using wavelets for feature extractions in tridesclous.