RPCA method in RAD doesn't work with multivariate data optimally

[mmolaro]

I haven't looked at the implementation details in the Pig code, but at least in the R interfaces, multivariate dataframes are treated one timeseries at a time.

This is not intuitive given the rational behind using RPCA for anomaly detection, in particular the existence of many dimensional datasets.

A data point that violates the lower dimensional subspace assumption by changing relationship between two variables could be easily missed in a one variable at a time perspective.