Cross-product matrices for STATIS (used to compute the table weights) become prohibitively large when high number of nodes per subject is used (e.g. 30k). It should be possible to reduce these matrices for inner product calculation, but this needs to be benchmarked against the original approach.