Advanced Machine Learning — Helping Organizations with Business Critical Data on IBM Z

Shutterstock image

For many years, the mainframe has held and processed some organizations’ most valued data due to its high levels of reliability, availability and security (RAS). The business-critical nature of this rich information source created a gravitational pull for advanced analytics solutions to be co-located with the data to help provide detailed insights, while maintaining the RAS capabilities organizations demand. This “data gravity” repositioned the mainframe as a high-performance hybrid transaction and analytics processing (HTAP) platform while removing unnecessary infrastructure latencies.

Machine Learning (ML) for z/OS was introduced in 2017. The intent of ML is to help organizations identify more opportunities for growth, reduce business costs and risks by building and deploying ML models that can continually learn as they encounter new data, helping predict outcomes faster, more often, with more consistency than humans alone.

In May 2018, ML for z/OS 1.2.0 was released delivering potentially more value to organizations including an enhanced and more advanced visual data exploration and model development tool via integration with IBM SPSS modeler catering to both coders and non-coders. Users can import existing SPSS Modeler streams into the SPSS Modeler integrated in the ML for z/OS UI (based on DSX Local). When creating a new SPSS Modeler flow, users can simply import the stream file with a couple of mouse clicks. The stream will render into the new SPSS Modeler UI. Users will be exposed to new interactive visualizations and have the ability to deploy the results in model management and deployment.

IBM DB2 Analytics Accelerator for z/OS (IDAA) works with ML for z/OS as a data source. With this integration, eligible SQL queries can be pushed to the Accelerator either via Jupyter notebooks or the Visual Modeler Builder of ML for z/OS. During model training, organizations have the benefit of analytic query acceleration while avoiding MIPS cost increase.

Finally, a higher performance scoring engine helps score very large models in less than 5 milliseconds[1].

For those interested in going a little deeper here’s a summary of what’s new in ML for z/OS 1.2.0:

• Integration of SPSS Modeler for developing more accurate models through a powerful, intuitive, graphical interface.
• Lifecycle management of PMML models that are trained in SPSS Modeler and RStudio.
• Interoperability of models between Machine Learning for z/OS and Data Science Experience Local.
• Availability of APIs for implementing Machine learning as a service for organizations.
• Audit trace capability for model governance.
• Performance enhancements in scoring engines, especially the scoring engine that runs in a CICS region.
• Security enhancements through the implementation of IzODA client authentication, user authentication with Knox and LDAP, and GDPR compliance requirements.
• Enhancements in scoring service management with z/OSMF workflow service and REST APIs.
• Enhancements in project and asset management and user collaboration.

Machine Learning Everywhere

In closing, ML for z/OS v1.2.0 is a good example of how IBM is helping organizations infuse ML and AI across their businesses. This release provides APIs for key machine learning capabilities including model deployment, scoring, evaluation, etc. These APIs help enable organizations embed these machine learning capabilities in existing applications helping to make them “smarter” — and embedding ML operationalization capabilities into current dev-ops processes.

To learn more about Machine Learning for z/OS 1.2.0 visit the IBM Knowledge Center

Steven Astorino, Vice President of Development, Private Cloud Platform and z Analytics

Follow me on twitter @astorino_steven

1. IBM internally tested an XGBoost model that was 1G large when converted to PMML with 2000 trees and 200+ input fields. The model was tested in an z/OS V2.2 environment with 1 GP and 4 zIIPs with 768 GB memory. Results showed that the round trip score for up to 16 threads averaged below 5 milliseconds.