---

name: datalab description: "Route end-to-end dataset auditing with cleanlab.Datalab." metadata: disco-role: operating disable-model-invocation: true license: Apache 2.0

datalab

Use this sub-skill when the user wants a unified dataset audit with cleanlab.Datalab.

Use it for

• One-pass auditing across multiple issue families.
• Datalab(...), find_issues(...), report(...), get_issues(...), and get_issue_summary(...).
• Classification, regression, and multilabel Datalab tasks.
• Optional image audits with image_key and CleanVision.
• Custom IssueManager workflows registered into Datalab.

Route elsewhere

• Standard noisy-label, CleanLearning, count, filter, rank, or dataset-health workflows → ../classification/SKILL.md.
• Standalone outlier scoring → ../outlier/SKILL.md.
• Multi-annotator consensus workflows → ../multiannotator/SKILL.md.
• Direct multilabel or regression label-issue workflows → ../tabular-label-issues/SKILL.md.
• Token, object-detection, or segmentation label issues → ../structured-label-issues/SKILL.md.
• Experimental deep-learning helpers or span classification → ../experimental/SKILL.md.

Core router facts

• Datalab accepts task="classification", task="regression", and task="multilabel".
• find_issues() accepts pred_probs, features, knn_graph, and issue_types.
• knn_graph wins when both knn_graph and features are provided.
• issue_types=None means use the task default set; issue_types={} means do nothing.
• report() prints to stdout and does not return a report string.
• get_issue_summary() is the dataset-level view; get_issues() is the per-example view.
• spurious_correlations is image-only and lives in get_info("spurious_correlations"), not in issue_summary.

Read/run next

• Read references/api-reference.md when you need live-verified constructor/method signatures, task values, or output table meanings.
• Read references/issue-types.md when deciding which issue families and inputs to request in find_issues().
• Read references/workflows.md when you need complete classification, regression, multilabel, kNN-graph, image, serialization, or report recipes.
• Read references/custom-issue-manager.md before registering a custom issue type.
• Read references/troubleshooting.md when dependencies, task names, issue_types, labels, images, or reports fail.
• Run scripts/smoke_datalab.py to verify a tiny Datalab audit and custom issue-manager path.