Prerequisites and Maintenance for Machine Learning for IBM z/OS

• Support multi-user model development environment leveraging JupyterHub.
• Enhanced trustworthy AI capability through improved monitoring configuration and intuitive visualization of explainability results.
• New capability in the Trustworthy AI component to continuously monitor and detect drift in the data distribution of model inputs and outputs, ensuring the model remains accurate and reliable.
• Support multi-class classification model scoring leveraging the on-chip AI accelerator of IBM z16 through Snap ML.
• Utilize Spark libraries as external dependencies for the scoring server to reduce the installation footprint.
• Support IBM Semeru Runtime Certified Edition for z/OS, Version 17, the latest release of Java SDK on z/OS.
• Support for z/OS Spark 3.5.0.
• Support IBM Z Deep Learning Compiler (IBM zDLC) 4.2.0 for ONNX model compilation and scoring.
• Provide the up to date machine learning support for AI Python packages on z/OS.
• Support multi-class classification model scoring leveraging the on-chip AI accelerator of IBM z16 through Snap ML.
• Utilize Spark libraries as external dependencies for the scoring server to reduce the installation footprint.
• Support IBM Semeru Runtime Certified Edition for z/OS, Version 17, the latest release of Java SDK on z/OS.
• New option of configuring a RACF® keyring-based keystore with hardware private key for user authentication and SSL/TLS communication
• Enhancement in support of issuing console messages to reflect when the core and scoring services have started, ended, or encountered an error
• Enhancement to drift-related output under Trustworthy AI
• Enhancement in support of integration with watsonx.ai and watsonx.governance
• Enhancement in support of online scoring timeout for PMML models
• Enhancement to Trustworthy AI in support of ONNX drift and exporting drift results as a PDF
• Dual control enhancement to enable an extra layer of security for online deployment changes
• Enhancement in support of Serving ID that enables inferencing using a user-provided ID in place of the auto-generated deployment ID
• Enhancement in support of Scoring service IP binding
• Support for the  z17 Telum II on-chip AI accelerator

What's New in version 3.1.0

• New ability to explain and understand the decisions made by AI models utilizing the MLz built-in Trustworthy AI capabilities.
• Integrate Snap ML library for high speed training and inferencing of popular machine learning models on z/OS, leveraging the on-chip AI accelerator of IBM z16.
• Provide the up-to-date machine learning support for Apache Spark and AI Python packages on z/OS.
• Improve the experience of product configuration and offer different configuration options through WMLz Configuration Tool for both WMLz Core edition and WMLz Enterprise edition.
• Incorporate the inferencing support for Watson Core time-series models.
• Provide the built-in metadata database option for repository service.
• Integrate the Jupyter Notebook server for model training to leverage the integrated Apache Spark runtime and Python runtime on z/OS.
• Enhance the documentation for WMLz services REST APIs.
• zIIP eligibility for Python model training and scoring.
• AT-TLS support for configuration tool, scoring service and Jupyter notebook server for strengthened TLS communication between WMLz services and your client applications.
• Enhanced error handling of the CICS and WOLA scoring interfaces, which enable COBOL applications to retrieve the scoring error code and error message through pre-defined output variables.
• Make the WOLA server name customizable, which enables a flexible deployment and application management for a scoring cluster configuration with instances across multiple LPARs.
• Message ID and content standardization for services and UI.
• Support IBM Z Deep Learning Compiler (IBM zDLC) 4.1.1 for ONNX model compilation and scoring.
• Support z/OS Spark 3.2.4.2.

Maintenance Level