Skip to main content
Open llms.txtCopy tools

CartoBoost Ranker

Use CartoBoostRanker when rows are only comparable within a query group. Typical uses are candidate ordering, route ranking, or planning contexts where only within-group comparisons matter.

The CartoBoost ranker uses pairwise logistic or LambdaRank objectives and reports grouped metrics from predictions.

Python Example​

from cartoboost import CartoBoostRanker

ranker = CartoBoostRanker(
    n_estimators=200,
    learning_rate=0.04,
    max_depth=4,
    split_policy="structured",
    objective="lambdarank",
)
ranker.fit(X_train, relevance_train, groups=query_sizes_train)
scores = ranker.predict(X_test)
metrics = ranker.score_groups(X_test, relevance_test, groups=query_sizes_test)

Browser WASM Example​

The browser bundle currently exposes the shared boosted-tree runner through runRegressionModel. Use this example to inspect WASM split behavior and model visualization; use the Python ranker API above when query groups and ranking metrics are required.

Boosted tree browser ranking analog

Runs runRegressionModel in Wasm with axis splitters and l2 loss.

Ready to run in this page.

Rows for each query must be contiguous. Pass groups as group sizes or contiguous query ids, or set group_col when the query id is a column in X.

Use When​

NeedBetter first choice
Rank candidates within each query.CartoBoostRanker
Predict an absolute score or amount.CartoBoostRegressor
Predict a class label or probability.CartoBoostClassifier
Forecast future time points.Forecasting models

Validation​

Use ranking metrics that match the decision: NDCG for graded relevance, MAP for retrieval-style relevance, and MRR when the first useful candidate matters most. Compare against simple popularity or dense tabular baselines under the same group split.

For workflow and method details, see Python API Reference.

Limitations​

  • Ranking labels are meaningful only within each query group.
  • Group boundaries must remain intact during splitting and prediction.
  • Offline ranking metrics do not by themselves establish downstream decision value.