Skip to contents

Multilabel Classification Metrics

Source: R/Metrics_multilabel.R
multilabel_metrics_fn.Rd

A configurable evaluator for multilabel classification tasks, where each sample can belong to multiple classes simultaneously. The function computes various discrimination metrics commonly used for multilabel problems.

Supported metrics names:

  • "accuracy" – Subset accuracy (exact match ratio)

  • "hamming_loss" – Hamming loss (fraction of incorrect labels)

  • "f1_micro" – F1 score, micro averaged across all labels

  • "f1_macro" – F1 score, macro averaged across labels

  • "f1_weighted" – F1 score, weighted averaged by label support

  • "f1_samples" – F1 score, averaged across samples

  • "precision_micro" – Precision, micro averaged

  • "precision_macro" – Precision, macro averaged

  • "precision_weighted" – Precision, weighted averaged

  • "precision_samples" – Precision, averaged across samples

  • "recall_micro" – Recall, micro averaged

  • "recall_macro" – Recall, macro averaged

  • "recall_weighted" – Recall, weighted averaged

  • "recall_samples" – Recall, averaged across samples

  • "jaccard_micro" – Jaccard index, micro averaged

  • "jaccard_macro" – Jaccard index, macro averaged

  • "jaccard_weighted" – Jaccard index, weighted averaged

  • "jaccard_samples" – Jaccard index, averaged across samples

  • "roc_auc_micro" – ROC AUC, micro averaged (requires y_prob)

  • "roc_auc_macro" – ROC AUC, macro averaged (requires y_prob)

  • "roc_auc_weighted" – ROC AUC, weighted averaged (requires y_prob)

  • "roc_auc_samples" – ROC AUC, averaged across samples (requires y_prob)

  • "pr_auc_micro" – PR AUC, micro averaged (requires y_prob)

  • "pr_auc_macro" – PR AUC, macro averaged (requires y_prob)

  • "pr_auc_weighted" – PR AUC, weighted averaged (requires y_prob)

  • "pr_auc_samples" – PR AUC, averaged across samples (requires y_prob)

Usage

multilabel_metrics_fn(y_true, y_prob, metrics = NULL, threshold = 0.5)

Arguments

y_true

Binary matrix or data frame of ground‑truth labels with shape (n_samples, n_labels), where 1 indicates presence and 0 indicates absence.

y_prob

Numeric matrix of predicted probabilities with shape (n_samples, n_labels). Will be converted to binary predictions using threshold.

metrics

Character vector listing which metrics to compute. Default is c("accuracy", "f1_micro", "f1_macro").

threshold

Numeric threshold for converting probabilities to binary predictions. Default is 0.5.

Value

Named numeric vector with one element per requested metric.

Examples

set.seed(42)
n <- 100
k <- 5
y_true <- matrix(rbinom(n * k, 1, 0.3), nrow = n, ncol = k)
y_prob <- matrix(runif(n * k), nrow = n, ncol = k)
multilabel_metrics_fn(y_true, y_prob, metrics = c("accuracy", "f1_micro"))
#>  accuracy  f1_micro 
#> 0.0400000 0.3264249