SLRM Model Settings

The Settings tab for a SLRM model nugget specifies options for modifying the built model. For example, you may use the SLRM node to build several different models using the same data and settings, then use this tab in each model to slightly modify the settings to see how that affects the results.

Note: This tab is only available after the model nugget has been added to a stream.

Maximum number of predictions per record. This option allows you to limit the number of predictions made for each record in the dataset. The default is 3.

For example, you may have six offers (such as savings, mortgage, car loan, pension, credit card, and insurance), but you only want to know the best two to recommend; in this case you would set this field to 2. When you build the model and attach it to a table, you would see two prediction columns (and the associated confidence in the probability of the offer being accepted) per record. The predictions could be made up of any of the six possible offers.

Level of randomization. To prevent any bias—for example, in a small or incomplete dataset—and treat all potential offers equally, you can add a level of randomization to the selection of offers and the probability of their being included as recommended offers. Randomization is expressed as a percentage, shown as decimal values between 0.0 (no randomization) and 1.0 (completely random). The default is 0.0.

Set random seed. When adding a level of randomization to selection of an offer, this option allows you to duplicate the same results in another session. By specifying the starting value used by the random number generator, you can ensure the same records are assigned each time the node is executed. Enter the desired seed value. If this option is not selected, a different sample will be generated each time the node is executed.

Note: When using the Set random seed option with records read from a database, a Sort node may be required prior to sampling in order to ensure the same result each time the node is executed. This is because the random seed depends on the order of records, which is not guaranteed to stay the same in a relational database.

Sort order. Select the order in which offers are to be displayed in the built model:

  • Descending. The model displays offers with the highest scores first. These are the offers that have the greatest probability of being accepted.
  • Ascending. The model displays offers with the lowest scores first. These are the offers that have the greatest probability of being rejected. For example, this may be useful when deciding which customers to remove from a marketing campaign for a specific offer.

Preferences for target fields. When building a model, there may be certain aspects of the data that you want to actively promote or remove. For example, if building a model that selects the best financial offer to promote to a customer, you may want to ensure that one particular offer is always included regardless of how well it scores against each customer.

To include an offer in this panel and edit its preferences, click Add, type the offer's name (for example, Savings or Mortgage), and click OK.

  • Value. This shows the name of the offer that you added.
  • Preference. Specify the level of preference to be applied to the offer. Preference is expressed as a percentage, shown as decimal values between 0.0 (not preferred) and 1.0 (most preferred). The default is 0.0.
  • Always include. To ensure that a specific offer is always included in the predictions, select this box.
    Note: If the Preference is set to 0.0, the Always include setting is ignored.

Take account of model reliability. A well-structured, data-rich model that has been fine-tuned through several regenerations should always produce more accurate results compared to a brand new model with little data. To take advantage of the more mature model's increased reliability, select this box.

Generate SQL for this model When using data from a database, SQL code can be pushed back to the database for execution, providing superior performance for many operations.

Select one of the following options to specify how SQL generation is performed.

  • Default: Score using Server Scoring Adapter (if installed) otherwise in process If connected to a database with a scoring adapter installed, generates SQL using the scoring adapter and associated user defined functions (UDF) and scores your model within the database. When no scoring adapter is available, this option fetches your data back from the database and scores it in SPSS® Modeler.
  • Score outside of the Database If selected, this option fetches your data back from the database and scores it in SPSS Modeler.