Run ML_PREDICT for XGBoost Classification & Regression Tasks - MaxCompute

ML_PREDICT runs inference on structured data using a pre-trained boosted tree model. Use it for classification and regression tasks directly in MaxCompute SQL.

Syntax

STRING|STRUCT ML_PREDICT(<model_name>, <version_name>, col1, col2, ...);

Parameters

Parameter	Required	Description
`model_name`	Yes	The name of the model. The model's `model_type` must be `BOOSTED_TREE_REGRESSOR` or `BOOSTED_TREE_CLASSIFIER`.
`version_name`	Yes	The model version to use. To use the default version, pass `DEFAULT_VERSION`.
`col1, col2, ...`	Yes	The input columns for prediction. The column type, order, and count must match the model definition. An error is returned if they do not match.

Return value

The return type depends on model_type:

Field	Type	Applies to	Description
(result)	`STRING`	`BOOSTED_TREE_REGRESSOR`	The predicted numeric value.
(result)	`STRUCT<label:STRING, probs:ARRAY(prob:FLOAT)>`	`BOOSTED_TREE_CLASSIFIER`	A struct containing the predicted class and per-class probabilities.
`label`	`STRING`	`BOOSTED_TREE_CLASSIFIER`	The predicted class for the row.
`probs`	`ARRAY(prob:FLOAT)`	`BOOSTED_TREE_CLASSIFIER`	The predicted probabilities for each class.

Examples

Regression model

Call the XGBoost regression model demo_model_xgboost_regressor at version v01. If this model does not exist, create it first. For details, see Train and predict with an XGBoost model in MaxCompute.

SELECT ML_PREDICT(demo_model_xgboost_regressor, v01, 0.22438f, 0f, 9.69f, 0, 0.585f, 6.027f, 79.7f, 2.4982f, 6.123f, 391f, 19.2f, 14.33f);

Result:

+----------+
| _c0      |
+----------+
| 20.03961 |
+----------+

Classification model

Call the XGBoost classification model test_xgboost_classifier at version v1. If this model does not exist, create it first. For details, see Create and delete a model.

SELECT ML_PREDICT(test_xgboost_classifier, v1, 1, 2, 3, 4);