The GBDT Binary Classification Prediction V2 component applies a trained gradient boosting decision tree (GBDT) model to new data and returns binary classification results. Use it downstream of the GBDT binary classification V2 training component, then connect the output to Binary classification evaluation to measure model performance.
Supported computing resources
This component runs on MaxCompute and Flink.
How it works
A GBDT model is an ensemble of decision trees, each acting as a weak learner. Trees are built sequentially: each tree optimizes the objective function on the basis of the previous decision tree, using a CART structure and a configurable step size. The final model combines all trees to produce predictions.
The figure below shows the basic recursive structure of gradient boosting.
In most cases, 
is a CART decision tree, 
are the parameters of the decision tree, and 
is the step size.
Configure the component in the PAI console
Input ports
Both input ports are required before the pipeline can run.
| Input port | Recommended upstream component | Required |
|---|---|---|
| Input | GBDT binary classification V2 | Yes |
| Predicted Data Table | Read Table | Yes |
Parameters
Fields Information tab
| Parameter | Required | Description | Default value |
|---|---|---|---|
| Prediction result column name | Yes | Name of the prediction result column. | prediction_result |
| predictionDetailCol | No | Name of the prediction details column. | prediction_detail |
| Reserved Columns | No | Columns from the input table to carry through to the output. All columns are carried through by default. | N/A |
Tuning tab
| Parameter | Required | Description | Default value |
|---|---|---|---|
| Number of Instances | No | Number of compute instances for the job. Increase this value when the job runs slowly on large datasets. | Auto-calculated based on input data |
| Memory Per Instance | No | Memory per instance, in MB. Valid values: 100–65536. Increase this value when the job fails with out-of-memory errors. | Auto-calculated based on input data |
Output ports
| Port | Recommended downstream component |
|---|---|
| Output | Binary classification evaluation |
Related topics
GBDT binary classification V2 — the upstream training component that produces the model used by this component.
Binary classification evaluation — evaluate prediction quality using the output of this component.
Designer overview — an introduction to Machine Learning Designer.
Overview of Designer components — the full list of available components.