The apt-transform maps a composition in the D-part real-simplex
linearly to a D-1 dimensional euclidian vector. Although the
transformation does not reach the whole \(R^{D-1}\), resulting covariance
matrices are typically of full rank.
The data can then
be analysed in this transformation by all classical multivariate
analysis tools not relying on distances. See
cpt
and ipt
for alternatives. The
interpretation of the results is easy since the relation to the first
D-1 original variables is preserved.
The additive planar transform is given by
$$ apt(x)_i := clo(x)_i, i=1,\ldots,D-1 $$