score_ylc: SCORE method from Yokota, Lee, and Cichoki (2017).
In dcgerard/hose: Higher-order spectral estimators

Description Usage Arguments Author(s) References

Implements the "Sparse CORE" method from Yokota, Lee, and Cichocki (2017). The idea is that one uses only a subset of the core array from the HOSVD to calculate the "modified" singular values. Then one uses the minimum description length (see mdl_eigen) criteria to choose the rank of the array. The proportion of the core should be small. We have one percent as the default as suggested in their paper, but my guess is that this should actually depend on the rank of the mean tensor.

1 2	score_ylc(Y, rho = 0.01, sampmin = min(c(dim(Y), 5)), return_est = TRUE, verbose = FALSE)

`Y`	An array of numerics.
`rho`	The proportion of the core array to use to calculate the mode-specific modified singular values.
`sampmin`	The minimum number of columns to of the core to use in calculating the mode-specific modified singular values.
`return_est`	A logical. Should we return the truncated HOSVD (`TRUE`) or not (`FALSE`)?
`verbose`	A logical. Should we print a lot?

David Gerard

Yokota, Tatsuya, Namgil Lee, and Andrzej Cichocki. "Robust multilinear tensor rank estimation using higher order singular value decomposition and information criteria." IEEE Transactions on Signal Processing 65.5 (2017): 1196-1206. DOI: 10.1109/TSP.2016.2620965

dcgerard/hose documentation built on Aug. 1, 2019, 12:11 a.m.