Multi scale features / targetting specific timepoints for improved classification + clustering

[abuzarmahmood]

Use additional shorter windows to differentiate spikes which have different "elbows"

Can also used Neighborhood component analysis or LDA to find what time-points in the spike-snapshot are MOST useful for separating waveforms.

[abuzarmahmood]

Weighted PCA using the below weights might be useful

[abuzarmahmood]

Using PCA of logistic regression coefficients (weighted transform) as transform improves classification accuracy

[abuzarmahmood]

[abuzarmahmood]

[abuzarmahmood]

[abuzarmahmood]

[abuzarmahmood]

Note : compression using this basis will only be useful for actual spikes and not noise