GBDT Binary Classification - Platform For AI - Alibaba Cloud Documentation Center

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 use one of the following methods to configure the GBDT Binary Classification component.

Method 1: Configure the component on the pipeline page

You can configure the parameters of the GBDT Binary Classification component on the pipeline page of Machine Learning Designer of Platform for AI (PAI). Machine Learning Designer is formerly known as Machine Learning Studio. The following table describes the parameters.

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.
Tuning	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.

Method 2: Use PAI commands

Configure the component parameters by using PAI commands. You can use the SQL Script component to call PAI commands. For more information, see SQL Script.

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 is 4: LOG_LIKELIHOOD.	4
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(Deprecated).	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. Unit: days.	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 a receiver operating characteristic (ROC) curve.

Example

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

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"

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";

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}