By Kewei Wei and Guanjun Cai
We just released IBM Watson Machine Learning for z/OS 2.2.0. This iteration of the industry-leading, end to end machine learning solution on IBM Z delivers many exciting new and enhanced capabilities based, in no small part, on your feedback.
This latest release represents our continuous effort to simplify the architecture of the solution, reducing its complexity and increasing its deployment flexibility. Depending on your machine learning need, you can now install the WMLz base on Z as a standalone solution, with the option to extend it by installing the WMLz IDE (integrated development environment) on Linux on Z or Linux systems.
In addition to architecture simplification, WMLz 2.2.0 also takes major strides towards model interoperability, service high availability, machine learning library customization, and ease of getting up-and-running:
Model interoperability. With this latest release, WMLz makes training anywhere and deploying anywhere a reality by keeping machine learning frameworks current and by providing workflows for cross-platform deployment.
Open source plays an instrumental role in the rapid development and adoption of the machine learning and AI technology. WMLz embraces open source by keeping up with the latest machine learning frameworks that are popular in the open source community. This release of WMLz supports Scikit-learn 0.22, XGBoost 0.90, and SparkML 2.4. As a result, your data scientist team can deploy in WMLz any model trained with the open source version of the same machine learning framework.
Besides keeping machine learning frameworks current, this release significantly improves the utilities and workflows for cross-platform model deployment. For example, your data scientists can train models on IBM Cloud Pak for Data, and your production engineers can click a button on the WMLz UI or make a simple REST call to import those models from a Cloud Pak for Data deployment space into WMLz for online scoring.
Last but not least, this release is shipped with new model utilities that enable you to easily save and import models trained locally on distributed platforms. Your data scientists can use the utilities to package both the content and the metadata of a model as a single compressed artifact, which makes it easy for model exporting, importing, deployment, and governance.
High availability of machine learning services. Making real-time predictions requires that machine learning services be highly available, reliable, and resilient. WMLz supports high availability first with the introduction of online scoring service in v1.2.0. This latest release extends that high availability architecture to all WMLz base core services.
The core services of WMLz base are for model training, deployment, batch scoring, ingestion, metadata repository, and data connection management. You can achieve high availability by configuring a WMLz base cluster. Each cluster can consist of two or more WMLz base instances that run either on a single LPAR or across different LPARs. To make core services in a WMLz base cluster highly available, reliable, and resilient is to keep one runtime environment active at all time.
A machine learning solution that leverages WMLz as the AI enabling infrastructure can benefit from this extended architecture of high availability. For example, when configured with this feature, IBM Db2 AI for z/OS that runs on the WMLz platform will be able to support mission-critical Db2 operations without the risk of a single point of failure in the WMLz core services.
Support of custom libraries. As machine learning and AI technology rapidly advances, more and more algorithms become available. But more often than not, data scientists find themselves in search for transformers and estimators that best serve their business need. They have to either constantly customize and enhance existing transformers and estimators or develop new ones all together.
In this latest release, WMLz enables data scientists to easily deploy the transformers or estimators they enhance or develop as custom libraries. After uploaded, the custom libraries can be managed along with the models that use them in WMLz. In addition, application developers can implement those custom transformers and estimators in different ways. WMLz can help suggest and choose the right implementation for the right context, including model training and scoring.
Guided configuration. Ease of onboarding, usability, and serviceability are three pillars of the WMLz goal of making machine learning and AI easy. This latest release continues to deliver on that goal by introducing a new configuration tool and enhancing the existing administration dashboard.
The new GUI configuration tool provides a one-stop service that guides you through the initial configuration of both the WMLz base and its prerequisite software on Z. The initial base configuration consists of verifying prerequisite software levels, verifying system environment variable settings, creating metadata repository in Db2 for z/OS, configuring keystore for secure connections and user authentication, creating z/OS Spark or Python runtime environments, and configuring WMLz UI and core services.
In addition to the guided configuration of WMLz base, this release also makes it easy for you to configure and manage scoring services by using the enhanced administration dashboard or the interactive shell script. You can easily create, register, update, start, stop, or remove a standalone or clustered scoring service that runs locally on the WMLz base server or remotely on a different server.
In short, WMLz 2.2.0 delivers new and enhanced functions, performance, and ease of use through its simplified architecture, reduced footprint, flexible deployment, guided configuration, core services high availability, custom libraries, and model interoperability, just to name a few. Together, all these new and enhanced capabilities will work to meet your machine learning need.
Learn more about IBM Watson Machine Learning for z/OS.
Kewei Wei (魏可伟), Senior Technical Staff Member, Lead Architect of IBM Watson Machine Learning for z/OS, IBM China Development Lab.
Guanjun Cai, Senior Information Architect and Developer of Watson Machine Learning for z/OS, Db2 for z/OS, and Db2 Data Gate, IBM Silicon Valley Lab.
Sign in and subscribe to always get the latest news about Db2 for z/OS from the IBM lab: http://ibm.biz/db2znews-subscribe