hosvd: Calculate the (truncated) higher-order SVD (HOSVD).

In tensr: Covariance Inference and Decompositions for Tensor Datasets

Description

Calculates the left singular vectors of each matrix unfolding of an array, then calculates the core array. The resulting output is a Tucker decomposition.

Usage

hosvd(Y, r = NULL)

Arguments

Y	An array of numerics.
r	A vector of integers. The rank of the truncated HOSVD.

Details

If r is equal to the rank of Y, then Y is equal to atrans(S, U), up to numerical accuracy.

More details on the HOSVD can be found in De Lathauwer et. al. (2000).

Value

U A list of matrices with orthonormal columns. Each matrix contains the mode-specific singular vectors of its mode.

S An all-orthogonal array. This is the core array from the HOSVD.

Author(s)

Peter Hoff.

References

De Lathauwer, L., De Moor, B., & Vandewalle, J. (2000). A multilinear singular value decomposition. SIAM journal on Matrix Analysis and Applications, 21(4), 1253-1278.

Examples

#Generate random data.
p <- c(2, 3, 4)
X <- array(stats::rnorm(prod(p)), dim = p)

#Calculate HOSVD.
hosvd_x <- hosvd(X)
S <- hosvd_x$S
U <- hosvd_x$U

#Recover X.
trim(X - atrans(S, U))

#S is all-orthogonal.
trim(mat(S, 1) %*% t(mat(S, 1)))
trim(mat(S, 2) %*% t(mat(S, 2)))
trim(mat(S, 3) %*% t(mat(S, 3)))

Example output

, , 1

     [,1] [,2] [,3]
[1,]    0    0    0
[2,]    0    0    0

, , 2

     [,1] [,2] [,3]
[1,]    0    0    0
[2,]    0    0    0

, , 3

     [,1] [,2] [,3]
[1,]    0    0    0
[2,]    0    0    0

, , 4

     [,1] [,2] [,3]
[1,]    0    0    0
[2,]    0    0    0

         [,1]     [,2]
[1,] 12.50138 0.000000
[2,]  0.00000 7.510044
         [,1]     [,2]     [,3]
[1,] 9.400672 0.000000 0.000000
[2,] 0.000000 6.730014 0.000000
[3,] 0.000000 0.000000 3.880738
         [,1]     [,2]     [,3]    [,4]
[1,] 9.769635 0.000000 0.000000 0.00000
[2,] 0.000000 5.538185 0.000000 0.00000
[3,] 0.000000 0.000000 3.078894 0.00000
[4,] 0.000000 0.000000 0.000000 1.62471

tensr documentation built on May 2, 2019, 2:32 p.m.