Within aspects of the present invention, correlation anomalies are
calculated in four stages for a reference data set, and a target data
set, respectively, the four stages are: Initially, a similarity matrix K
of the variables to be diagnosed is calculated. This similarity matrix is
embedded in a low-dimensional vector space in order to calculate and
obtain the coordinates {z.sub.1, . . . z.sub.p} for each variable. Next,
an energy value {e.sub.1, . . . e.sub.p} is calculated per variable from
calculated coordinates. Lastly, a comparison is made of the energy values
that have been calculated for the target data set with the energy values
of that have been calculated for the reference data set in order to
determine the degree of correlation anomalies that has occurred between
the two data sets.