Step 6: Configure Prediction Parameters

  • Article

You can modify the tables used by the Predictor resource to customize prediction parameters.

In this lab, you will learn how to modify the PredictorAttributes_Demographics table.

The PredictorAttributes_Demographics table is created when you build a new model for the first time. The table is empty with no rows and all the default values are enabled. A row in this table contains either a change in the attributes of one property in a dense table, or a change in the attributes of all properties in a sparse table.

  1. Click Start, point to Programs, point to Microsoft SQL Server, and then click Enterprise Manager.
  2. Expand Microsoft SQL Servers, expand SQL Server Group, expand <server name>, expand Databases, expand PredictorDemo_dw, and then click Tables.
  3. From the Tables list, right-click the PredictorAttributes_Demographics table, point to Open Table, and then click Return all Rows.
  4. In the SQL Server Enterprise Manager – [Data in Table ‘PredictorAttributes_Demographics' in ‘PredictorDemo_dw' on ‘<server name>'] window, click the last row marked by *, and then do the following:
    Use this To do this
    PropID Leave this column blank.

    This column holds a sequential number and will be filled in automatically
    ParentID Type –1.

    This column will let the Predictor resource change the attributes of a property or the entire table.
    Name Type Age.

    This column will let the Predictor resource change the attributes of the property Age.
    TableName Leave this column blank.
    ColumnName Leave this column blank.
    Distribution Type 1.

    A null value (if Distribution left blank) tells the system to guess the distribution type for the property Age. In this example, the system guesses that the distribution type for Age is continuous, model as binary.

    Value 1 tells the system to use distribution type normal instead, not model as binary.

    The flag model as binary indicates that the only important distinction to be made about a property is whether it is present or missing. Often, transactional properties are modeled as binary (for example, products purchased are modeled as binary when it matters only to know if a product was purchased or not), whereas demographic properties should usually not be set that way. (The system incorrectly guesses model as binary for Age in this example because there are many missing values.)

    The possible distribution types are:

    • 0: Discrete, Not Model As Binary,
    • 1: Normal, Continuous, Not Model As Binary
    • 2: LogNormal, Continuous, Not Model As Binary
    • 3: not valid
    • 4: Discrete, Model As Binary
    • 5: Normal, Continuous, Model As Binary
    • 6: Lognormal, Continuous, Model As Binary
    • 7: not valid
    • 8: model as advertisement
    • <Null>: Autodetect (the system guesses for you)
    UseToPredict Type 1.

    This selection will let the Predictor resource consider Age when building decision trees for other properties.

    The value UseToPredict is typically set to 1, but you can set it to 0 to prevent the property from being used for prediction. Setting the value to 0 is useful when you know the property will not be available for predictions.
    Predict Type 1 to have the Predictor resource build a decision tree for predicting Age.

    If this column is set to 0, a decision tree will not be built.
  5. Close the SQL Server Enterprise Manager.

Copyright © 2005 Microsoft Corporation.
All rights reserved.