6. Discretized Partial Least Squares
(dPLS)
Introduction
In this vignette, we consider approximating multiple matrices as a product of ternary (or non-negative) low-rank matrices (a.k.a., factor matrices).
Test data is available from toyModel.
You will see that there are five blocks in the data matrix as follows.
suppressMessages(library("fields"))
layout(t(1:3))
image.plot(X[[1]], main="X1", legend.mar=8)
image.plot(X[[2]], main="X2", legend.mar=8)
image.plot(X[[3]], main="X3", legend.mar=8)
Semi-Ternary Simultaneous Matrix Factorization (STSMF)
Here, we introduce the ternary regularization to take {-1,0,1} values in Vk as below:
max tr(Vj′Xj′XkVk) s.t. j ≠ k, V ∈ {−1, 0, 1},
where j and k range from 1 to K, K is the number of matrices, Xk (N × Mk)
is a k-th data matrix and
Vk (Mk × J)
is a k-th ternary loading
matrix. In dcTensor package, the object function is
optimized by combining gradient-descent algorithm (Tsuyuzaki 2020) and ternary regularization.
Basic Usage
In STSMF, a rank parameter J ( ≤ min (N, M)) is needed to
be set in advance. Other settings such as the number of iterations
(num.iter) are also available. For the details of arguments
of dPLS, see ?dPLS. After the calculation, various objects
are returned by dPLS. STSMF is achieved by specifying the
ternary regularization parameter as a large value like the below:
## List of 6
## $ U :List of 3
## ..$ : num [1:100, 1:3] 8722 8926 8821 8626 8589 ...
## ..$ : num [1:100, 1:3] 5888 5898 6044 5910 5695 ...
## ..$ : num [1:100, 1:3] 3879 3904 3961 3806 3909 ...
## $ V :List of 3
## ..$ : num [1:300, 1:3] 0.96 0.966 0.973 0.95 0.925 ...
## ..$ : num [1:200, 1:3] 0.887 0.892 0.881 0.913 0.913 ...
## ..$ : num [1:150, 1:3] 0.0252 0.0332 0.0286 0.0346 0.0244 ...
## $ RecError : Named num [1:101] 1.00e-09 1.88e+06 1.87e+06 1.83e+06 1.79e+06 ...
## ..- attr(*, "names")= chr [1:101] "offset" "1" "2" "3" ...
## $ TrainRecError: Named num [1:101] 1.00e-09 1.88e+06 1.87e+06 1.83e+06 1.79e+06 ...
## ..- attr(*, "names")= chr [1:101] "offset" "1" "2" "3" ...
## $ TestRecError : Named num [1:101] 1e-09 0e+00 0e+00 0e+00 0e+00 0e+00 0e+00 0e+00 0e+00 0e+00 ...
## ..- attr(*, "names")= chr [1:101] "offset" "1" "2" "3" ...
## $ RelChange : Named num [1:101] 1.00e-09 9.92e-01 5.11e-03 2.20e-02 2.12e-02 ...
## ..- attr(*, "names")= chr [1:101] "offset" "1" "2" "3" ...The reconstruction error (RecError) and relative error
(RelChange, the amount of change from the reconstruction
error in the previous step) can be used to diagnose whether the
calculation is converged or not.
layout(t(1:2))
plot(log10(out_dPLS$RecError[-1]), type="b", main="Reconstruction Error")
plot(log10(out_dPLS$RelChange[-1]), type="b", main="Relative Change")
The products of Uk and Vk (k = 1…K) show whether the
original data matrices are well-recovered by dPLS.
recX <- lapply(seq_along(X), function(x){
out_dPLS$U[[x]] %*% t(out_dPLS$V[[x]])
})
layout(rbind(1:3, 4:6))
image.plot(t(X[[1]]))
image.plot(t(X[[2]]))
image.plot(t(X[[3]]))
image.plot(t(recX[[1]]))
image.plot(t(recX[[2]]))
image.plot(t(recX[[3]]))
The histograms of Vks show that all the factor matrices Vk looks ternary.
layout(rbind(1:3, 4:6))
hist(out_dPLS$U[[1]], breaks=100)
hist(out_dPLS$U[[2]], breaks=100)
hist(out_dPLS$U[[3]], breaks=100)
hist(out_dPLS$V[[1]], breaks=100)
hist(out_dPLS$V[[2]], breaks=100)
hist(out_dPLS$V[[3]], breaks=100)
Session Information
## R version 4.4.1 (2024-06-14)
## Platform: x86_64-pc-linux-gnu
## Running under: Ubuntu 24.04.1 LTS
##
## Matrix products: default
## BLAS: /usr/lib/x86_64-linux-gnu/openblas-pthread/libblas.so.3
## LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/libopenblasp-r0.3.26.so; LAPACK version 3.12.0
##
## locale:
## [1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C
## [3] LC_TIME=en_US.UTF-8 LC_COLLATE=C
## [5] LC_MONETARY=en_US.UTF-8 LC_MESSAGES=en_US.UTF-8
## [7] LC_PAPER=en_US.UTF-8 LC_NAME=C
## [9] LC_ADDRESS=C LC_TELEPHONE=C
## [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C
##
## time zone: Etc/UTC
## tzcode source: system (glibc)
##
## attached base packages:
## [1] stats graphics grDevices utils datasets methods base
##
## other attached packages:
## [1] nnTensor_1.3.0 fields_16.2 viridisLite_0.4.2 spam_2.10-0
## [5] dcTensor_1.3.0 rmarkdown_2.28
##
## loaded via a namespace (and not attached):
## [1] gtable_0.3.5 jsonlite_1.8.8 highr_0.11 compiler_4.4.1
## [5] maps_3.4.2 Rcpp_1.0.13 plot3D_1.4.1 tagcloud_0.6
## [9] jquerylib_0.1.4 scales_1.3.0 yaml_2.3.10 fastmap_1.2.0
## [13] ggplot2_3.5.1 R6_2.5.1 tcltk_4.4.1 knitr_1.48
## [17] MASS_7.3-61 dotCall64_1.1-1 misc3d_0.9-1 tibble_3.2.1
## [21] maketools_1.3.0 munsell_0.5.1 pillar_1.9.0 bslib_0.8.0
## [25] RColorBrewer_1.1-3 rlang_1.1.4 utf8_1.2.4 cachem_1.1.0
## [29] xfun_0.47 sass_0.4.9 sys_3.4.2 cli_3.6.3
## [33] magrittr_2.0.3 digest_0.6.37 grid_4.4.1 rTensor_1.4.8
## [37] lifecycle_1.0.4 vctrs_0.6.5 evaluate_0.24.0 glue_1.7.0
## [41] buildtools_1.0.0 fansi_1.0.6 colorspace_2.1-1 pkgconfig_2.0.3
## [45] tools_4.4.1 htmltools_0.5.8.1