sequential selection can result in problems with eigendecomposition

[jacobdenobel]

I think this happens only when used in combination with active. Probably, the negative weight multiplication results in nan values in the covariance matrix update.