Robust Distances producing NaN and Inf

Hello Friends,

I have a distribution of multivariate data and am using the Mahalanobis distance (MD) as a metric for some decision making. I am trying to calculate the MD a couple of different ways to assess performance. More specifically, I am comparing the MD when sample covariance and sample mean are used vs. more robust estimations of scatter and location using MCD and OGK.

When I calculate the MD using sample mean and covariance, my results are good. When I calculate using robustcov, 'method', 'ogk' everything is good. But when I use fastMCD (robustcov(X)) I get a bunch of Inf's and Nan's.

I don't really understand why I would be getting these singularities. Any ideas?

-Mike

[Image Analyst]

Including your data and lines of code would certainly help you get an answer.

Answer 1

[Walter Roberson]

Crystal ball says...

You used / instead of ./