AdaBoost Example 1 - Teradata Vantage

This example uses home sales data to create a model that predicts home style, which can be input to the AdaBoostPredict Example 1.

Input

• AttributeTable: housing_train_att, created by inputting housing_train (the raw data for DecisionForest Example 1: Classification Tree without Out-of-Bag Error) to the Unpivoting function:

  CREATE MULTISET TABLE housing_train_att AS (
    SELECT * FROM Unpivoting (
      ON housing_train
      USING
      TargetColumns ('price', 'lotsize', 'bedrooms', 'bathrms',
        'stories', 'driveway', 'recroom', 'fullbase', 'gashw', 'airco',
        'garagepl', 'prefarea')
      Accumulate ('sn')
    ) AS dt
  ) WITH DATA;

• CategoricalAttributeTable: housing_cat, created and populated with this code:

  CREATE MULTISET TABLE housing_cat (attribute VARCHAR);
  
  INSERT INTO housing_cat VALUES ('driveway');
  INSERT INTO housing_cat VALUES ('recroom');
  INSERT INTO housing_cat VALUES ('fullbase');
  INSERT INTO housing_cat VALUES ( 'gashw');
  INSERT INTO housing_cat VALUES ('airco');
  INSERT INTO housing_cat VALUES ('prefarea');

• ResponseTable: housing_train_response, created by selecting the columns sn and homestyle from housing_train:

  CREATE MULTISET TABLE housing_train_response AS (
    SELECT sn, homestyle AS response
    FROM housing_train
  ) WITH DATA;

This query returns the following table:

SELECT * FROM housing_train_att ORDER BY 1, 2;

This query returns the following table:

SELECT * FROM housing_cat ORDER BY 1;

This query returns the following table:

SELECT * FROM housing_train_response ORDER BY 1;

SQL Call

This call creates the model, abd_model, using the default values for the optional arguments.

SELECT * FROM AdaBoost (
  ON housing_train_att AS AttributeTable
  ON housing_cat AS CategoricalAttributeTable
  ON housing_train_response AS ResponseTable
  OUT TABLE OutputTable (abd_model)
  USING
  IdColumns ('sn')
  AttributeNameColumns ('attribute')
  AttributeValueColumn ('value_col')
  ResponseColumn ('response')
  IterNum (20)
  NumSplits (10)
  MaxDepth (3)
  MinNodeSize (100)
) AS dt ;

Output

Because the argument IterNum has the value 20, the function builds 20 classification trees.

Running 20 round AdaBoost, computing 20 classification trees.
AdaBoost model created.

This query returns the following table:

SELECT * FROM abd_model ORDER BY 1,3;