Tensor Toolbox (MATLAB)

Work-in-progress

This page of the docs is still a work-in-progress. Check back later!

Translation table

Tensor Toolbox	This package
`NORMALIZE(X)`	`normalizecomps(X)`
`NORMALIZE(X,N)`	`normalizecomps(X; distribute_to=N)`
`NORMALIZE(X,0)`	`normalizecomps(X; distribute_to=1:ndims(X))`
`NORMALIZE(X,[])`	`normalizecomps(X)`
`NORMALIZE(X,'sort')`	`sortcomps!(normalizecomps(X))`
`NORMALIZE(X,N,1)`	`normalizecomps(X, 1; distribute_to=N)`
`NORMALIZE(X,0,1)`	`normalizecomps(X, 1; distribute_to=1:ndims(X))`
`NORMALIZE(X,[],1)`	`normalizecomps(X, 1)`
`NORMALIZE(X,'sort',1)`	`sortcomps!(normalizecomps(X, 1))`
`NORMALIZE(X,[],1,I)`	`normalizecomps(X, 1; dims=I)`
`NORMALIZE(X,[],2,I)`	`normalizecomps(X, 2; dims=I)`
`X = ARRANGE(X)`	`sortcomps!(normalizecomps!(X))`
`X = ARRANGE(X,N)`	`sortcomps!(normalizecomps!(X; distribute_to=N); dims=N, by=norm)`
`X = ARRANGE(X,P)` where `P` is a permutation	`permutecomps!(X, P)`
`X = REDISTRIBUTE(X,N)`	`normalizecomps!(X; dims=:λ, distribute_to=N)`
`B = PERMUTE(A,ORDER)`	`B = permutedims(A, ORDER)`

Noteworthy differences

`normalizecomps` v.s. `NORMALIZE` in Tensor Toolbox

normalizecomps does not fix the signs of the weights to be positive.
normalizecomps supports
- normalizing any subset of the weight vector and factor matrices,
- with respect to any $\ell_p$ norm, and
- distributing the excess weight into any subset of the weight vector and factor matrices.
normalizecomps comes in two varieties:
- normalizecomps! normalizes the CPD in-place.
- normalizecomps makes a new copy.

Ref: https://gitlab.com/tensors/tensor_toolbox/-/blob/v3.6/@ktensor/normalize.m