---

name: classification description: "Route standard multiclass and binary noisy-label workflows for cleanlab." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

classification

Use this sub-skill for standard single-label binary or multiclass workflows centered on noisy labels, cross-validated predicted probabilities, label-quality scoring, class-level dataset health, latent/noise-matrix estimation, and synthetic noise benchmarks.

Route here for

• CleanLearning fit/predict/get_label_issues workflows.
• Direct label-issue finding or ranking from pred_probs.
• Cross-validated predicted probabilities and latent noise estimation.
• Dataset health summaries, class overlap, and overall label-quality scores.
• Data valuation with data_shapley_knn.
• Benchmark noise generation with the synthetic noise helpers.

Route elsewhere for

• Dataset audits that span multiple issue types -> ../datalab/SKILL.md.
• Outlier or OOD scoring from features or pred_probs alone -> ../outlier/SKILL.md.
• Multiannotator consensus, annotator quality, or active learning -> ../multiannotator/SKILL.md.
• Multi-label or regression label issues -> ../tabular-label-issues/SKILL.md.
• Token, object-detection, or segmentation label issues -> ../structured-label-issues/SKILL.md.
• Explicitly unstable/deep-learning experimental helpers, including the batched low-memory helper itself -> ../experimental/SKILL.md.

Operating rules

• Labels must be zero-based integers for the standard multiclass/binary route.
• pred_probs must be out-of-sample, aligned to the same examples, and ordered by class index.
• If you do not already have out-of-sample pred_probs, compute them with cross-validation first.
• Keep the classifier sklearn-compatible and clonable when using CleanLearning.
• If the user wants a full dataset audit, do not force this route; hand off to datalab.

Read/run next

• Read references/api-reference.md when you need signatures, return shapes, or function ownership for the classification APIs.
• Read references/workflows.md when you need a concrete pred_probs, CleanLearning, dataset-health, noise-estimation, or benchmarking recipe.
• Read references/data-valuation.md when the task asks for KNN-Shapley/data-valuation scores.
• Read references/troubleshooting.md when imports, labels, pred_probs, sklearn estimator compatibility, or cross-validation fail.
• Run scripts/smoke_classification.py to verify a tiny deterministic classification, dataset-health, noise-generation, and data-valuation path.