Scoring

Scoring is the process of generating real-time values by supplying predictive models with input data. A scoring model is any artifact that can be used to produce output values given input data, such as a PMML file from IBM® SPSS® Statistics.

In general, to use a model for generating scores:

Select a model to use for scoring from the IBM SPSS Collaboration and Deployment Services Repository.
Define a scoring configuration for the model.
Supply the configured model with data and generate scores.

The predictive model used for scoring can be defined using IBM SPSS Modeler streams or PMML that is generated from IBM Corp. products. IBM SPSS Collaboration and Deployment Services supports PMML 4.2 and earlier versions. In addition, legacy markup (e.g. SPSS-ML) from older products may also be used for scoring. Note that some nodes in streams, such as the ADP node, must be trained before they can be used for scoring. For more information, see the IBM SPSS Modeler documentation.

Important: To use a particular model type for scoring, a scoring adapter for that model type must be installed on the IBM SPSS Collaboration and Deployment Services Repository server. For example, to generate scores based on PMML files, the IBM SPSS Collaboration and Deployment Services Scoring Adapter for PMML must be installed. To use IBM SPSS Modeler files for scoring, the IBM SPSS Collaboration and Deployment Services Scoring Adapter for IBM SPSS Modeler must be installed.

Use the IBM SPSS Deployment Manager to define model scoring configurations and to monitor the model scoring performance. Results generated from scoring can be viewed in the IBM SPSS Collaboration and Deployment Services Deployment Portal or in custom client applications.

Figure 1. General scoring architecture

Figure 1 illustrates the general architecture for scoring. In this example, the IBM SPSS Collaboration and Deployment Services Repository contains three predictive models. Configuration A defines the scoring settings for Model 1. Configuration B defines the scoring settings for Model 2. Configurations C and D define different scoring settings for Model 3.

IBM SPSS Collaboration and Deployment Services Deployment Portal users send separate scoring requests to models 1 and 3 that use the settings in configurations A and C. The scoring service generates the scores and returns the values to the users.

As an alternative to using IBM SPSS Collaboration and Deployment Services Deployment Portal, you can create a custom scoring client to send scoring requests and to receive results. In this example, a custom client sends separate scoring requests to models 2 and 3 that use the settings in configurations B and D. The scoring service generates the scores and returns the values to the client.