TODO: describe what it looks like for dimensions to match up between
s and mask. See vignette("sparse-computations") for mathematical
details.
masked_approximation_impl(U, V, row, col)A sparse matrix representing the low-rank reconstruction
from U, d and V, only at the index pairs indicated by
row and col.
Low-rank matrix of left singular-ish vectors.
Low-rank matrix of right singular-ish vectors.
Zero-based row indices of observed elements.
Zero-based col indices of observed elements.
The idea is to populate U, d and V with using the
elements of an SVD-like list. You can generate row and col
most easily from a sparse masking Matrix (Matrix package),
coercing to triplet format, and extracting mask@i for row
and mask@j for column.