Closed cgranade closed 7 years ago
In lieu of #109 being accepted, I'm merging #109 into this PR to enable Travis CI testing.
This PR now passes all local and Travis tests, so in principle it should be ready to merge in. That said, I want to do some more "applied" tests and make sure that it works at least in the example that prompted #108.
It seems that in some cases, eigenvalues that are negative due to numerical instability (~10⁻¹⁵ – 10⁻¹⁶) can cause errors. I'll fix and amend the PR accordingly.
Thanks to #116, it looks as though this PR now passes and should be ready to merge once it gets a review.
Awesome, thanks for reviewing, I'll merge it in now!
This PR should fix #108 by adding a more stable algorithm for LW covariance estimation.