The GBDT Binary Classification component is used to set a threshold. If a feature value is greater than the threshold, the feature is a positive example. Otherwise, the feature is a negative example.

Configure the component

You can configure the component by using one of the following methods:
  • Use the Machine Learning Platform for AI console
    Tab Parameter Description
    Fields Setting Feature Columns The feature columns that are selected from the input table for training. The columns of the DOUBLE and BIGINT types are supported.
    Note A maximum of 800 feature columns can be selected.
    Label Column The label column. Only the column of the BIGINT type is supported.
    Stratified Column The columns of the DOUBLE and BIGINT types are supported. By default, the full table is a group.
    Parameters Setting Metric Type The metric type. Valid values: NDCG and DCG.
    Decision Tree Quantity The number of trees. Valid values: 1 to 10000.
    Learning Rate The learning rate. Valid values: (0,1).
    Rate of Samples for Training The proportion of samples that are selected for training. Valid values: (0,1].
    Ratio of Features for Training The proportion of features that are selected for training. Valid values: (0,1].
    Maximum Leaf Quantity The maximum number of leaf nodes on each tree. Valid values: 1 to 1000.
    Testing Data Ratio The proportion of data that is selected for testing. Valid values: [0,1).
    Maximum Decision Tree Depth The maximum depth of each tree. Valid values: 1 to 100.
    Minimum Number of Samples on a Leaf Node The minimum number of samples on each leaf node. Valid values: 1 to 1000.
    Random Seed The random seed. Valid values: [0,10].
    Maximum Feature Split Times The maximum number of splits of each feature. Valid values: 1 to 1000.
    Tuning Number of Cores The number of cores. The system automatically allocates cores used for training based on the volume of input data.
    Memory The memory size of each core. The system automatically allocates the memory based on the volume of input data. The memory size of each core. Valid values: 1024 to 64 × 1024. Unit: MB.
  • Use commands
    PAI -name gbdt_lr
        -project algo_public
        -DfeatureSplitValueMaxSize="500"
        -DrandSeed="0"
        -Dshrinkage="0.5"
        -DmaxLeafCount="32"
        -DlabelColName="y"
        -DinputTableName="bank_data_partition"
        -DminLeafSampleCount="500"
        -DgroupIDColName="nr_employed"
        -DsampleRatio="0.6"
        -DmaxDepth="11"
        -DmodelName="xlab_m_GBDT_LR_21208"
        -DmetricType="2"
        -DfeatureRatio="0.6"
        -DinputTablePartitions="pt=20150501"
        -DtestRatio="0.0"
        -DfeatureColNames="age,previous,cons_conf_idx,euribor3m"
        -DtreeCount="500"
    Parameter Required Description Default value
    inputTableName Yes The name of the input table. N/A
    featureColNames No The feature columns that are selected from the input table for training. All columns of numeric data types
    labelColName Yes The name of the label column in the input table. N/A
    inputTablePartitions No The partitions that are selected from the input table for training. Specify this parameter in one of the following formats:
    • Partition_name=value
    • name1=value1/name2=value2: multi-level partitions
    Note If you specify multiple partitions, separate these partitions with commas (,).
    All partitions
    modelName Yes The name of the output model. N/A
    outputImportanceTableName No The name of the table that provides feature importance. N/A
    groupIDColName No The name of the group column. Full table
    lossType No The type of the loss function. Valid values:
    • 0: GBRank
    • 1: LAMBDAMART_DCG
    • 2: LAMBDAMART_NDCG
    • 3: LEAST_SQUARE
    • 4: LOG_LIKELIHOOD
    0
    metricType No The metric type. Valid values:
    • 0: normalized discounted cumulative gain (NDCG).
    • 1: discounted cumulative gain (DCG).
    • 2: area under the curve (AUC). This metric type is suitable only for the scenario where the value of label is set to 0 or 1.
    0
    treeCount No The number of trees. Valid values: 1 to 10000. 500
    shrinkage No The learning rate. Valid values: (0,1). 0.05
    maxLeafCount No The maximum number of leaf nodes on each tree. Valid values: 1 to 1000. 32
    maxDepth No The maximum depth of each tree. Valid values: 1 to 100. 10
    minLeafSampleCount No The minimum number of samples on each leaf node. Valid values: 1 to 1000. 500
    sampleRatio No The proportion of samples that are selected for training. Valid values: (0,1]. 0.6
    featureRatio No The proportion of features that are selected for training. Valid values: (0,1]. 0.6
    tau No The Tau parameter for the GBRank loss function. Valid values: [0,1]. 0.6
    p No The p parameter for the GBRank loss function. Valid values: [1,10]. 1
    randSeed No The random seed. Valid values: [0,10]. 0
    newtonStep No Specifies whether to use Newton's method. Valid values: 0 and 1. 1
    featureSplitValueMaxSize No The maximum number of splits of each feature. Valid values: 1 to 1000. 500
    lifecycle No The lifecycle of the output table. N/A
Note
  • By default, the loss function types of gradient boosting decision tree (GBDT) and those of GBDT_LR are different. By default, GBDT uses regression loss:mean squared error loss as its loss function, and GBDT_LR uses logistic regression loss as its loss function. Therefore, you do not need to configure a loss function for GBDT_LR.
  • The feature column, label column, and stratification column of GBDT must be of numeric data types.
  • You must specify the objective reference value for the Prediction component to generate an receiver operating characteristic (ROC) curve.

Example

  1. Execute the following SQL statements to generate training data:
    drop table if exists gbdt_lr_test_input;
    create table gbdt_lr_test_input
    as
    select
        *
    from
    (
        select
            cast(1 as double) as f0,
            cast(0 as double) as f1,
            cast(0 as double) as f2,
            cast(0 as double) as f3,
            cast(0 as bigint) as label
        from dual
        union all
            select
                cast(0 as double) as f0,
                cast(1 as double) as f1,
                cast(0 as double) as f2,
                cast(0 as double) as f3,
                cast(0 as bigint) as label
        from dual
        union all
            select
                cast(0 as double) as f0,
                cast(0 as double) as f1,
                cast(1 as double) as f2,
                cast(0 as double) as f3,
                cast(1 as bigint) as label
        from dual
        union all
            select
                cast(0 as double) as f0,
                cast(0 as double) as f1,
                cast(0 as double) as f2,
                cast(1 as double) as f3,
                cast(1 as bigint) as label
        from dual
        union all
            select
                cast(1 as double) as f0,
                cast(0 as double) as f1,
                cast(0 as double) as f2,
                cast(0 as double) as f3,
                cast(0 as bigint) as label
        from dual
        union all
            select
                cast(0 as double) as f0,
                cast(1 as double) as f1,
                cast(0 as double) as f2,
                cast(0 as double) as f3,
                cast(0 as bigint) as label
        from dual
    ) a;
    The following training data table gbdt_lr_test_input is generated.
    f0 f1 f2 f3 label
    1.0 0.0 0.0 0.0 0
    0.0 0.0 1.0 0.0 1
    0.0 0.0 0.0 1.0 1
    0.0 1.0 0.0 0.0 0
    1.0 0.0 0.0 0.0 0
    0.0 1.0 0.0 0.0 0
  2. Run the following PAI command to submit the training parameters configured for the GBDT Binary Classification component:
    drop offlinemodel if exists gbdt_lr_test_model;
    PAI -name gbdt_lr
        -project algo_public
        -DfeatureSplitValueMaxSize="500"
        -DrandSeed="1"
        -Dshrinkage="1"
        -DmaxLeafCount="30"
        -DlabelColName="label"
        -DinputTableName="gbdt_lr_test_input"
        -DminLeafSampleCount="1"
        -DsampleRatio="1"
        -DmaxDepth="10"
        -DmodelName="gbdt_lr_test_model"
        -DmetricType="0"
        -DfeatureRatio="1"
        -DtestRatio="0"
        -DfeatureColNames="f0,f1,f2,f3"
        -DtreeCount="5"
  3. Run the following PAI command to submit the parameters configured for the Prediction component:
    drop table if exists gbdt_lr_test_prediction_result;
    PAI -name prediction
        -project algo_public
        -DdetailColName="prediction_detail"
        -DmodelName="gbdt_lr_test_model"
        -DitemDelimiter=","
        -DresultColName="prediction_result"
        -Dlifecycle="28"
        -DoutputTableName="gbdt_lr_test_prediction_result"
        -DscoreColName="prediction_score"
        -DkvDelimiter=":"
        -DinputTableName="gbdt_lr_test_input"
        -DenableSparse="false"
        -DappendColNames="label";
  4. View the prediction result table gbdt_lr_test_prediction_result.
    label prediction_result prediction_score prediction_detail
    0 0 0.9984308925552831 {"0": 0.9984308925552831, "1": 0.001569107444716943}
    0 0 0.9984308925552831 {"0": 0.9984308925552831, "1": 0.001569107444716943}
    1 1 0.9982721832240973 {"0": 0.001727816775902724, "1": 0.9982721832240973}
    1 1 0.9982721832240973 {"0": 0.001727816775902724, "1": 0.9982721832240973}
    0 0 0.9984308925552831 {"0": 0.9984308925552831, "1": 0.001569107444716943}
    0 0 0.9984308925552831 {"0": 0.9984308925552831, "1": 0.001569107444716943}