Logistic regression classifier for factor embeddings

Trains a multinomial logistic regression on factor embeddings using stats::glm (one-vs-rest for > 2 classes) and evaluates on test samples with the same comprehensive metrics as classify_embedding.

Usage

classify_logistic(
  embedding,
  labels,
  test_fraction = 0.2,
  test_idx = NULL,
  seed = NULL
)

Arguments

embedding

Numeric matrix where rows are samples and columns are features (e.g., t(result$H) for sample embeddings).

labels

Integer or factor vector of class labels. Length must equal nrow(embedding).

test_fraction

Fraction of samples held out for testing (default 0.2).

test_idx

Optional integer vector of test indices. If provided, test_fraction is ignored.

seed

Random seed for train/test split reproducibility.

Value

An fn_classifier_eval object (same structure as KNN variant).

See also

classify_embedding, classify_rf

Examples

# \donttest{
# Logistic regression on random embeddings
set.seed(42)
embed <- matrix(rnorm(200), nrow = 40, ncol = 5)
labels <- factor(rep(1:4, each = 10))
eval <- classify_logistic(embed, labels, seed = 1)
eval$accuracy
#> [1] 0.5
# }