---

name: online-core-api description: "Explain River's estimator lifecycle, base interfaces, cloning, tags, and estimator validation checks." disable-model-invocation: true metadata: disco-role: operating license: BSD 3-Clause

online-core-api

Use this sub-skill when a task asks how a River object should learn, predict, transform, clone, expose tags, or pass generic estimator checks.

Read first

• references/estimator-contracts.md
• references/api-reference.md
• references/troubleshooting.md
• scripts/estimator_contract_smoke.py --help

Use this for

• Explaining learn_one, predict_one, predict_proba_one, transform_one, score_one, update, and mini-batch counterparts.
• Adding or reviewing a River estimator, transformer, clusterer, anomaly detector, drift detector, or wrapper.
• Deciding whether a model is supervised, multiclass, stochastic, or tag-sensitive.
• Using clone, mutate, _unit_test_params, _unit_test_skips, and checks.check_estimator.
• Diagnosing why a classifier emits None or {} before seeing labels.

Route elsewhere

• Pipeline composition, selectors, feature extraction, preprocessing, and feature-routing patterns belong to pipelines-and-features.
• Datasets, stream adapters, progressive validation, and metrics loops belong to streaming-evaluation.
• Choosing a supervised model family belongs to supervised-models.
• Drift, anomaly, clustering, time-series, bandit, and recommender workflows belong to specialized-workflows.

Core decision points

• Use one-sample methods unless the task explicitly names mini-batch APIs or pandas dataframes.
• Treat dictionaries as the canonical feature container; missing and new feature keys are normal in River.
• Prefer generic checks for new estimator compatibility, then add model-family-specific tests for behavior not covered by generic checks.
• If a pipeline is checked with isinstance(pipeline, base.Classifier), River unwraps the last step for estimator-type checks.