Specify Model Settings

Understand how to configure machine learning models at build time.

Numerous configuration settings are available for configuring machine learning models at build time. To specify settings, create a settings table with the columns shown in the following table and pass the table to CREATE_MODEL.

You can use CREATE_MODEL2 procedure where you can directly pass the model settings to a variable that can be used in the procedure. The variable can be declared with DBMS_DATA_MINING.SETTING_LIST procedure.

Table 4-1 Settings Table Required Columns

Column Name	Data Type
`setting_name`	`VARCHAR2(30)`
`setting_value`	`VARCHAR2(4000)`

Example 4-1 creates a settings table for a Support Vector Machine (SVM) classification model. Since SVM is not the default classifier, the ALGO_NAME setting is used to specify the algorithm. Setting the SVMS_KERNEL_FUNCTION to SVMS_LINEAR causes the model to be built with a linear kernel. If you do not specify the kernel function, the algorithm chooses the kernel based on the number of attributes in the data.

Example 4-2 creates a model with the model settings that are stored in a variable from SETTING_LIST.

Some settings apply generally to the model, others are specific to an algorithm. Model settings are referenced in Table 4-2 and Table 4-3.

Table 4-2 General Model Settings

Settings	Description
Machine learning function settings	Machine Learning Technique Settings
Algorithm names	Algorithm Names
Global model characteristics	Global Settings
Automatic Data Preparation	Automatic Data Preparation

Table 4-3 Algorithm-Specific Model Settings

Algorithm	Description
CUR Matrix Decomposition	DBMS_DATA_MINING —Algorithm Settings: CUR Matrix Decomposition
Decision Tree	DBMS_DATA_MINING —Algorithm Settings: Decision Tree
Expectation Maximization	DBMS_DATA_MINING —Algorithm Settings: Expectation Maximization
Explicit Semantic Analysis	DBMS_DATA_MINING —Algorithm Settings: Explicit Semantic Analysis
Exponential Smoothing	DBMS_DATA_MINING —Algorithm Settings: Exponential Smoothing Models
Generalized Linear Model	DBMS_DATA_MINING —Algorithm Settings: Generalized Linear Models
k-Means	DBMS_DATA_MINING —Algorithm Settings: k-Means
Multivariate State Estimation Technique - Sequential Probability Ratio Test	DBMS_DATA_MINING - Algorithm Settings: Multivariate State Estimation Technique - Sequential Probability Ratio Test
Naive Bayes	Algorithm Settings: Naive Bayes
Neural Network	DBMS_DATA_MINING —Algorithm Settings: Neural Network
Non-Negative Matrix Factorization	DBMS_DATA_MINING —Algorithm Settings: Non-Negative Matrix Factorization
O-Cluster	Algorithm Settings: O-Cluster
Random Forest	DBMS_DATA_MINING — Algorithm Settings: Random Forest
Singular Value Decomposition	DBMS_DATA_MINING —Algorithm Settings: Singular Value Decomposition
Support Vector Machine	DBMS_DATA_MINING —Algorithm Settings: Support Vector Machine
XGBoost	DBMS_DATA_MINING — Algorithm Settings: XGBoost

Note:

Some XGBoost objectives apply only to classification function models and other objectives apply only to regression function models. If you specify an incompatible objective value, an error is raised. In the DBMS_DATA_MINING.CREATE_MODEL procedure, if you specify DBMS_DATA_MINING.CLASSIFICATION as the function, then the only objective values that you can use are the binary and multi values. The one exception is binary: logitraw, which produces a continuous value and applies only to a regression model. If you specify DBMS_DATA_MINING.REGRESSION as the function, then you can specify binary: logitraw or any of the count, rank, reg, and survival values as the objective.

The values for the XGBoost objective setting are listed in the Settings for Learning Tasks table in DBMS_DATA_MINING — Algorithm Settings: XGBoost.

Example 4-1 Creating a Settings Table and Creating an SVM Classification Model Using CREATE.MODEL procedure

CREATE TABLE svmc_sh_sample_settings (
  setting_name VARCHAR2(30),
  setting_value VARCHAR2(4000));

BEGIN 
  INSERT INTO svmc_sh_sample_settings (setting_name, setting_value) VALUES
    (dbms_data_mining.algo_name, dbms_data_mining.algo_support_vector_machines);
  INSERT INTO svmc_sh_sample_settings (setting_name, setting_value) VALUES
    (dbms_data_mining.svms_kernel_function, dbms_data_mining.svms_linear);
  COMMIT;
END;
/
-- Create the model using the specified settings 
BEGIN
  DBMS_DATA_MINING.CREATE_MODEL(
    model_name          => 'svm_model',
    mining_function     => dbms_data_mining.classification,
    data_table_name     => 'mining_data_build_v',
    case_id_column_name => 'cust_id',
    target_column_name  => 'affinity_card',
    settings_table_name => 'svmc_sh_sample_settings');
END;

Example 4-2 Specify Model Settings for a GLM Regression Model Using CREATE_MODEL2 procedure

DECLARE
    v_setlist DBMS_DATA_MINING.SETTING_LIST;
BEGIN
    v_setlist('PREP_AUTO') := 'ON';
    v_setlist('ALGO_NAME') := 'ALGO_GENERALIZED_LINEAR_MODEL';
    v_setlist('GLMS_DIAGNOSTICS_TABLE_NAME') := 'GLMR_DIAG';
    v_setlist('GLMS_FTR_SELECTION') := 'GLMS_FTR_SELECTION_ENABLE';
    v_setlist('GLMS_FTR_GENERATION') := 'GLMS_FTR_GENERATION_ENABLE';
 
    DBMS_DATA_MINING.CREATE_MODEL2(
        MODEL_NAME          => 'GLM_REGR',
        MINING_FUNCTION     => 'REGRESSION',
        DATA_QUERY          => 'select * from TRAINING_DATA',
        SET_LIST            => v_setlist,
        CASE_ID_COLUMN_NAME => 'HID',
	TARGET_COLUMN_NAME  => 'MEDV');
END;

Related Topics

Oracle Database PL/SQL Packages and Types Reference

Model Settings

Oracle Machine Learning uses settings to specify the algorithm and model settings or hyperparameters. Some settings are general, some are specific to a machine learning function, and some are specific to an algorithm.

OML4SQL