limited the amount of data copying when passing between methods
added regularization term to logDet(M) for numerical stability. The small (1e-10) negative eigenvalues of the separate matrices would accumulate into the final additive M due to round off error. Thus, det(M)<0 and the log would blow up, returning nan.
logDet(M)for numerical stability. The small (1e-10) negative eigenvalues of the separate matrices would accumulate into the final additiveMdue to round off error. Thus,det(M)<0and thelogwould blow up, returningnan.lMax==-1after searching.