mahalanobis distance with non symmetric and positive covariance

I have an array of 36 vectors with 46 dimensions. I'm trying to find the distance between my 36 vectors by using mahalanobis distance. I can't apply it in matlab because the covariance result is not symmetric and positive definite.

Could you help me with that please ?

Answer 1

[John D'Errico]

You can always use my nearestSPD tool, as posted on the file exchange.

A covariance matrix based on fewer samples than variables must always be singular. I'm not sure what this says about the distances you would compute, but nearestSPD will make a matrix usable for any tool that requires a Cholesky decomposition.

Answer 2

[feras almasri]

Thanks a lot that helped. But I think the result is not appropriate to use the mahalanobis distance cause the matrix result was wrong.