Align a document-term matrix to a grid of fitted models

Every OpTop tool requires the document-term matrix and the models to share one vocabulary in one order, because probabilities are aligned by term name and never by position. When they drift apart (a feature was trimmed after fitting, an engine reordered its vocabulary, models in the grid were fit on slightly different preprocessing), the index functions stop and point here. This helper subsets and reorders the columns of dtm to the intersection of the models' vocabularies, in the term order of the first model. Recompute the partition and the baseline on the aligned matrix afterwards; they encode the support.

Usage

optop_align_dtm_to_models(dtm, models)

Arguments

dtm

A document-term matrix of counts (rows are documents).

models

A list of fitted topic models supported by optop_as_theta_phi(), including optop_model() objects.

Value

The realigned document-term matrix, with columns ordered as the common vocabulary.

Details

Accepts a quanteda::dfm (returned as dgCMatrix) or any matrix with column names.

See also

optop_make_partition() and optop_make_baseline(), which must be recomputed on the aligned matrix.

Examples

rdirich <- function(n, k) {
  g <- matrix(stats::rgamma(n * k, shape = 1), n, k)
  g / rowSums(g)
}
theta <- rdirich(4, 2)
rownames(theta) <- sprintf("doc%d", 1:4)
phi <- rdirich(2, 6)
colnames(phi) <- sprintf("w%02d", 1:6)
m <- optop_model(theta, phi)

# counts whose columns are shuffled and carry one extra feature
counts <- matrix(rpois(4 * 7, 3), 4, 7,
                 dimnames = list(rownames(theta),
                                 c("w03", "extra", "w01", "w06",
                                   "w02", "w05", "w04")))
aligned <- optop_align_dtm_to_models(counts, list(m))
identical(colnames(aligned), m$terms)
#> [1] TRUE