emperor
emperor copied to clipboard
Add co variance matrix as an option in --ellipsoid_method
@rob-knight suggested in #39 to add an option to calculate the co variance matrix to determine the dimensions of the ellipsoids when jackknifing.
The co variance matrix should give you a tilted oval modeling the dataset as multivariate normal. Yes, each pc is a variate, matrix holds cov(i,j) with var(i) on diagonal, is used to calculate ellipsoid.
@antgonza found an available implementation of this in numpy, see:
http://docs.scipy.org/doc/numpy/reference/generated/numpy.cov.html
This gist includes the source code that Ryan Kennedy shared with Rob: https://gist.github.com/ElDeveloper/31d95bf37ac5aed8070f
Once implemented, this should be the default.