CREATE MODEL

Function

CREATE MODEL trains a machine learning model and saves the model.

Precautions

• The model name must be unique. Pay attention to the naming format.
• The AI training duration fluctuates greatly, and in some cases, the training duration is long. If the duration specified by the GUC parameter statement_timeout is too long, the training will be interrupted. You are advised to set statement_timeout to 0 so that the statement execution duration is not limited.

Syntax

CREATE MODEL model_name USING algorithm_name 
[FEATURES { {expression [ [ AS ] output_name ]} [, ...] }]
[TARGET { {expression [ [ AS ] output_name ]} [, ...] }]
FROM { table_name | select_query }
WITH hyperparameter_name = { hyperparameter_value | DEFAULT } [, ...] }

Parameter Description

• model_name

  Name of the training model, which must be unique.

  Value range: a string. It must comply with the identifier naming convention.

• architecture_name

  Algorithm type of the training model.

  Value range: a string. Currently, the value can be logistic_regression, linear_regression, svm_classification, or kmeans.

• attribute_list

  Enumerated input column name of the training model.

  Value range: a string. It must comply with the naming convention of data attributes.

• attribute_name

  Target column name of the retraining model in a supervised learning task (simple expression processing can be performed).

  Value range: a string. It must comply with the naming convention of data attributes.

• subquery

  Data source.

  Value range: a string. It must comply with the SQL syntax of databases.

• hyper_parameter_name

  Hyperparameter name of the machine learning model.

  Value range: a string. The value range varies depending on the algorithms. For details, see Table 2.

• hp_value

  Hyperparameter value.

  Value range: a string. The value range varies depending on the algorithms. For details, see Table 3.

Examples

CREATE MODEL price_model USING logistic_regression
 FEATURES size, lot
 TARGET price
 FROM HOUSES
 WITH learning_rate=0.88, max_iterations=default;

Helpful Links

DROP MODEL and PREDICT BY