All Products
Search

This Product

    Platform For AI:XGBoost train

    Document Center

    Platform For AI:XGBoost train

    Last Updated:Mar 11, 2026

    Train gradient boosting models for classification and regression tasks using XGBoost algorithm.

    PAI extends the open-source XGBoost library with distributed computing support and platform integration.

    Compute resources

    Supported compute resources:

    • MaxCompute

    • Flink

    • Deep Learning Containers (DLC)

    Data formats

    Accepted input formats: Table or LibSVM. Specify through the Field Setting tab: featureCols for Table format, vectorCol for LibSVM format (mutually exclusive).

    Table format

    Each row represents one training instance. Columns store feature values and label.

    f0 f1 label
    0.1 1 0
    0.9 2 1

    LibSVM format

    Each row contains sparse feature-value pairs in index:value format. Example:

    2:1 9:1 10:1 20:1 29:1 33:1 35:1 39:1 40:1 52:1 57:1 64:1 68:1 76:1 85:1 87:1 91:1 94:1 101:1 104:1 116:1 123:1
0:1 9:1 18:1 20:1 23:1 33:1 35:1 38:1 41:1 52:1 55:1 64:1 68:1 76:1 85:1 87:1 91:1 94:1 101:1 105:1 115:1 121:1
2:1 8:1 18:1 20:1 29:1 33:1 35:1 39:1 41:1 52:1 57:1 64:1 68:1 76:1 85:1 87:1 91:1 94:1 101:1 104:1 116:1 123:1
2:1 9:1 13:1 21:1 28:1 33:1 36:1 38:1 40:1 53:1 57:1 64:1 68:1 76:1 85:1 87:1 91:1 94:1 97:1 105:1 113:1 119:1
0:1 9:1 18:1 20:1 22:1 33:1 35:1 38:1 44:1 52:1 55:1 64:1 68:1 76:1 85:1 87:1 91:1 94:1 101:1 104:1 115:1 121:1
0:1 8:1 18:1 20:1 23:1 33:1 35:1 38:1 41:1 52:1 55:1 64:1 68:1 76:1 85:1 87:1 91:1 94:1 101:1 105:1 116:1 121:1

    Configuration

    Configure the component using three tabs:

    • Field Setting: Input data structure and model storage location

    • Parameter Setting: XGBoost training algorithm hyperparameters

    • Execution Tuning: Distributed training resources

    Field Setting

    Parameter Type Description
    labelCol String Label column name.
    featureCols String array Feature columns in Table format. Mutually exclusive with vectorCol. Requires Table-formatted input data.
    vectorCol String Vector column name in LibSVM format. Mutually exclusive with featureCols. Requires LibSVM-formatted input data.
    weightCol String Weight column name.
    set the model file path String OSS bucket path for storing the trained model.

    Parameter Setting

    Standard XGBoost hyperparameters.

    Training objective

    Parameter Type Default Description
    The number of rounds for boosting Integer -- Number of boosting iterations.
    objective String binary:logistic Learning task and corresponding learning objective.
    Base score Floating-point number 0.5 Global bias used as initial prediction score for all instances.
    The number of classes Integer -- Number of classes. Required for multi-class classification.

    Tree construction

    Parameter Type Default Description
    Tree Method String auto Tree construction algorithm. Valid values: auto, exact, approx, hist.
    Maximum depth of a tree Integer 6 Maximum tree depth. Higher values increase model complexity and overfitting risk.
    Max leaves Integer 0 Maximum number of leaf nodes. 0 means no limit.
    Grow Policy String depthwise Node addition strategy. Valid values: depthwise, lossguide.
    Maximum number of discrete bins to bucket continuous features Integer 256 Maximum number of discrete bins for continuous features. Applies only when Tree Method is hist.
    sketch_eps Floating-point number 0.03 Accuracy of binning in the sketching algorithm. Applies only when Tree Method is approx.

    Regularization

    Parameter Type Default Description
    L1 regularization term on weights Floating-point number 0.0 L1 regularization on weights. Higher values produce more conservative models.
    L2 regularization term on weights Floating-point number 1.0 L2 regularization on weights. Higher values produce more conservative models.
    gamma Floating-point number 0.0 Minimum loss reduction required for further leaf node partition.
    Min child weight Floating-point number 1.0 Minimum sum of instance weights required in a child node.
    Max delta step Floating-point number 0.0 Maximum delta step per leaf node. Controls model update granularity.

    Learning rate and sampling

    Parameter Type Default Description
    eta Floating-point number 0.3 Learning rate. Lower values improve robustness but require more boosting rounds.
    scale_pos_weight Floating-point number 1.0 Weight ratio for balancing positive and negative classes. Useful for imbalanced datasets.
    Subsample ratio of the training instances Floating-point number 1 Fraction of training instances sampled before each boosting round.
    Sampling method String GRADIENT_BASED Sampling method for training instances. Valid values: GRADIENT_BASED, UNIFORM.
    Subsample ratio of columns for each level Floating-point number 1.0 Fraction of columns sampled per tree level.
    Subsample ratio of columns for each node (split) Floating-point number 1.0 Fraction of columns sampled per node split.
    Subsample ratio of columns when constructing each tree Floating-point number 1.0 Fraction of columns sampled per tree.

    Constraints and distribution

    Parameter Type Default Description
    Interaction constraints String -- Feature groups allowed to interact.
    Monotone constraints String -- Monotonicity constraints per feature.
    Tweedie variance power Floating-point number 1.5 Variance power of the Tweedie distribution. Applies only when objective uses Tweedie distribution.

    Execution Tuning

    Parameter Type Valid values Description
    Number of Workers Positive integer [1, 9999] Number of worker nodes. Configure with Memory per worker, unit MB.
    Memory per worker, unit MB Positive integer [1024, 64 x 1024] Memory allocated per worker node, in MB.

    Usage example

    This example demonstrates XGBoost classification on Higgs boson event data using a preset template. To create this pipeline from the Use XGBoost algorithm to identify the Higgs boson template, see Create a pipeline from a preset template.

    Convert the output format for evaluation

    The component outputs JSON strings serialized from the open-source XGBoost library. Convert this output to enable evaluation components to process predictions.

    Add an SQL Script component downstream of XGboost Predict. The SQL Script converts predictions into the format required by Binary classification Evaluation V2. Use this code:

    set odps.sql.udf.getjsonobj.new=true;

select *, CONCAT("{\"0\":", 1.0-prob, ",\"1\":", prob, "}") as detail
FROM (
select *, cast(get_json_object(pred, '$[0]') as double) as prob FROM ${t1})

    References