Integrating a Machine Learning model

To call a Machine Learning model from your rule project, you must import the model into Rule Designer.

Before you begin

You must have the Miniloan decision service and the Mortgage Approval Prediction Model downloaded to your computer. See Applying a Machine Learning model.

About this task

This task shows you how to add a Machine Learning model to a decision service. You import the model into Rule Designer and integrate it into the decision service.

Procedure

  1. Import the Miniloan decision service into Rule Designer:
    1. Open Rule Designer on your computer.
    2. In the Rule perspective, click File > Import.
    3. In the Import wizard, expand General > Existing Projects into Workspace, and click Next.
    4. Choose Select root directory, and browse to the Miniloan files that you downloaded and expanded on your computer.
    5. Click Select All to pick all the Miniloan files.
    6. In Options, click Copy projects into workspace.
    7. Click Finish.
    The Rule perspective now shows the Miniloan decision service.
  2. Right-click the Miniloan project and select Machine Learning Integration > Enhance with Machine Learning.
    The selection dialog for ODM Machine Learning references opens.
  3. Select Create to import a Machine Learning model.
    The Add a Machine Learning Call panel opens.
  4. Under Machine Learning data, click Browse and import the JSON or YAML file for the Mortgage Approval Prediction Model.
    After you import the model file, most of the fields in the Add a Machine Learning Call panel are populated with default values from the file.
  5. Add a deployment ID to the ML Deployment ID field. Use the deployment ID that you obtained in step 18 of Downloading a Machine Learning model.
    The deployment ID is a unique model ID that you obtain from the Machine Learning user interface. If you do not have this ID, enter a placeholder value instead. You can change the value later by using the Update button in the selection dialog.
  6. Review the default values and correct invalid values.
    The values might be invalid if you used them in a previous Machine Learning import or the characters in the filename are not accepted values.
  7. Click Close.
    You can now modify the decision service to use the Machine Learning model.

Results

Importing the model creates the following elements in Rule Designer:
  • XOM with two Java™ classes:
    • Class 1:
      • Name taken from the bottom field in the create panel
      • Getters and setters for the input values in the swagger file
      • Method to set the deployment ID (setDeploymentID(String))
      • Method to do a model execution (execute())
      • Method to query if the returned values are from the defaults (isReturningDefaultValues())
      • Method to get the a return code from the call (getReturnCode()). (List below of possible return codes.)
      • Getters for the output values
    • Class 2:

      If Machine Learning cannot be accessed at run time, you can configure it to use default values by specifying that this is permitted in your rules. When default values are needed, the runtime obtains them from this bean class.

  • A BOM class is created that provides a verbalization for the new XOM
  • A variable set is created with two variables:
    • BOM class to be used for accessing Machine Learning
    • Deployment ID for use when calling Machine Learning
  • Sample rules in the rule package:
    • Rule 1 calls Machine Learning.
    • Rule 2 uses the score from Machine Learning.

      This sample rules can be used in your call. At the end, the rule prints out the returned values. You delete these print statements and use the output values in the rules that follow in the ruleflow.

What to do next

In the next task, you modify a decision service to use a Machine Learning model in Rule Designer. Then, you publish the decision service to Decision Center and deploy it to Rule Execution Server for testing.