Skip to content

using_pm_service.md

Latest commit

 

History

History
188 lines (145 loc) · 5.65 KB

using_pm_service.md

File metadata and controls

188 lines (145 loc) · 5.65 KB
copyright
years
2016, 2017
lastupdated 2017-08-09

{:new_window: target="_blank"} {:shortdesc: .shortdesc} {:screen: .screen} {:codeblock: .codeblock} {:pre: .pre}

Using the service

The modeling methods available on the SPSS Modeler modeling palette enable you to derive new information from your data and to develop predictive models. Each method has certain strengths and is best suited for particular types of problems. {: .shortdesc}

For details about SPSS Modeler and the modeling algorithms it provides, see IBM Knowledge Center.

After the input and output requirements of your Bluemix application and SPSS Modeler scoring branch design are implemented, your Data Analyst can change any internal aspect of the scoring branch. The Data Analyst can even change the model algorithm(s) used in a refresh operation, ensuring your ability to fine-tune your predictive analytics without needing to rewrite your applications.

Note the following important information regarding the Machine Learning service when used with models created in SPSS Modeler:

  • When a scoring branch is prepared for use in real-time scoring, input data coming in on the score request must replace the source node designed into the scoring branch and the resulting predictive analytic output must flow back into the response flow (effectively replacing the terminal node in the scoring branch design).

  • As the scoring branch is prepared for real-time execution in Bluemix, it cannot require a connection to an external service. For example, an IBM Analytical Decision Management scoring branch design cannot contain references to rules or models stored in an IBM SPSS Collaboration and Deployment Services repository.

  • The execution of a scoring branch for real-time scoring in Bluemix cannot require an external service. For example, you cannot deploy and score against model algorithms that require an IBM SPSS Analytic Server and Apache Hadoop data store in real-time.

  • Machine Learning supports Modeler embedded Python scripting. There are a few restrictions due to the method used for processing streams before they run in Machine Learning. Typically, if a user chooses to control the execution of the stream, they will reference the terminal node of the branch. For Machine Learning, when we process the stream, we identify the nodes from JSON that will be overridden, and then do the replacement before the stream runs. This causes the stream to fail in the script because the referenced input and export nodes no longer exist. The solution is use the ID of another node to uniquely identify the branch during execution. This ensures that the stream executes as defined in the embedded Python script.

For more details about current support for IBM SPSS Analytic Server-trained predictive models, see the Analytic Server section of IBM Knowledge Center.

Complete the following steps to create your Bluemix application and bind it to the Machine Learning service.

  1. Download Node.js sample application code from github repository.

  2. Use the cf create-service command to create a service instance:

    cf create-service pm-20 Free {local naming}

    {: codeblock}

    For example:

    cf create-service pm-20 Free my_pm_free

    {: codeblock}

    This command creates one Machine Learning service instance with Free plan named my_pm_free in your Bluemix space.

  3. Use the cf create-service-key command to create service credentials:

    cf create-service-key "{service instance name}" {vcap key name}

    {: codeblock}

    For example:

    cf create-service-key "IBM Watson Machine Learning - my instance" Credentials-1

    {: codeblock}

    This command creates Machine Learning service credentials.

  4. Use the cf bind-service command to bind the service instance my_pm_free to your application.

    cf bind-service AppName my_pm_service

    {: codeblock}

    For example:

    cf bind-service my_app1 my_pm_free

    {: codeblock}

    This command binds the Machine Learning service instance my_pm_free to the Bluemix application my_app1.

  5. Machine Learning credentials:

    After you bind the Machine Learning service instance to your Bluemix application, the Machine Learning credentials are added to the VCAP_SERVICES environment variable:

    {   
        "pm-20": {      
            "name": "pm20-1",
            "label": "pm-20",
            "plan": "Free",
            "credentials": {
                "url": "https://ibm-watson-ml.mybluemix.net",
                "access_key": "XXXXXXXXXXXXX"
            }
        }       
    }

{: codeblock}

The VCAP_SERVICES environment variable includes the following information:

plan
The Machine Learning plan that is used in the service provisioning.
url
The address of the Machine Learning service instance.
access_key
The query parameter accessKey to pass in all requests to this service instance.

For example:

Get https://ibm-watson-ml.mybluemix.net/pm/v1/model/sales_model2?accesskey=XXXXXXXXXXXXX

{: codeblock}

Example Node.js code that shows how to obtain the accessKey from the VCAP_SERVICES environment variable:

   if (process.env.VCAP_SERVICES) {
        var env = JSON.parse(process.env.VCAP_SERVICES);
        var credentials = env['pm-20'][0].credentials;
        var accessKey = credentials.access_key;
    }

{: codeblock}