Forest_Predict Example - Teradata Vantage

Input

SQL Call

Use the Accumulate argument to pass the homestyle variable, to easily compare the actual and predicted response for each observation.

CREATE MULTISET TABLE rf_housing_predict AS (
  SELECT * FROM Forest_Predict@coprocessor (
    ON housing_test PARTITION BY ANY
    ON rft_model AS ModelTable DIMENSION
    USING
    NumericInputs ('price ', 'lotsize ', 'bedrooms ', 'bathrms ',
                   'stories ', 'garagepl')
    CategoricalInputs ('driveway ', 'recroom ', 'fullbase ', 'gashw ',
                       'airco ', 'prefarea')
    IDColumn ('sn')
    Accumulate ('homestyle')
    Detailed ('false')
  ) AS dt
) WITH DATA;

Output

The predicted response is in the prediction column and the original classification values are in the homestyle column. The upper and lower confidence intervals are also shown in the output table.

This query returns the following table:

SELECT * FROM rf_housing_predict ORDER BY 2;

Prediction Accuracy

This query returns the prediction accuracy:

SELECT (SELECT count(sn) FROM rf_housing_predict
  WHERE homestyle = prediction) / (SELECT count(sn)
    FROM rf_housing_predict) AS PA;