Machine Learning

Release 10.15.0

Improvements in Machine Learning Engine

• Enhancements:
  • In the Machine Learning web application, confirmation popups have been modified to provide a consistent UX with the rest of the platform.

• Upgrades:

  • Machine Learning web application upgraded to use Cumulocity IoT Web SDK 1015.0.2.

  • Zementis microservice upgraded from openssl version 1.0.2 to version 1.1.1n.

  • Nyoka and Onnx microservices upgraded from NumPy version 1.18.5 to version 1.21.5.

  • Nyoka and Onnx microservices upgraded from scikit-learn version 0.21.3 to version 0.23.1.

  • Onnx microservice upgraded from ONNX version 1.7.0 to version 1.11.0.

  • Onnx microservice upgraded from ONNX-runtime version 1.5.2 to version 1.11.0.

  • Onnx microservice upgraded from protobuf version 3.17.0 to version 3.20.0.

• Bug fixes:
  • Fixed the issue where requests to the Onnx microservice were failing if user agent validation was enabled in the OAI settings.
  • Fixed the issue where the model group list was displaying the deleted model groups occasionally.
  • Fixed the issue where listed pmml models reported “404-Not Found” while attempting to delete them.

Improvements in Machine Learning Workbench

• Compute off-load with a pluggable Jupyter Enterprise Gateway (JEG):
  With this release, the Cumulocity IoT Machine Learning Workbench provides an option to connect to a Jupyter Enterprise Gateway which enables launch and management of kernels on behalf of remote notebooks. The JEG exposes a kernel-as-a-service model providing better optimization of compute resources, improved multi-user support, enable compute offload capability and more granular security for Jupyter notebook environment.

  • Settings:
    • A new tab is added to register the JEG URL
    • Popup to notify the user that all running kernels will be deleted during the setting and delete actions of the JEG credentials
  • Notebook UI:
    • Popup display to select different kernel specs
    • “Jupyter” logo beside the kernel name on every notebook to notify the user that the notebook is running using a kernel spec from the remote JEG server
    • Option within the integrated Jupyter UI to insert the code snippets which allow a data scientist to:
      • Download “Data” from MLW to the JEG server
      • Upload “Models” to MLW from the JEG server
      • List the projects available
      • List the resources available within a project

• MLW SDK:
  To complement the inclusion of the Jupyter Enterprise Gateway connection within the Cumulocity IoT Machine Learning Workbench, we also provide a Python package which facilitates exchange of artifacts (Data, Models, Resources) to and from the JEG server.

• Enhancements:
  • Project management UX improvements:

    • Ability to select multiple resources for deletion

    • Ability to delete an individual project version

    • Ability to select multiple projects for deletion

    • Task and asset grouping based on projects

    • Ability to select multiple tasks and assets for deletion

    • Project commit button relocated to make the functionality more intuitive

• Upgrades:

  • Machine Learning Workbench web application upgraded to use Cumulocity IoT Web SDK 1015.0.2.

  • MLW microservice upgraded from ONNX version 1.7.0 to version 1.11.0.

  • MLW microservice upgraded from protobuf version 3.12.4 to version 3.20.0.

  • MLW microservice upgraded from supervisor version 4.2.1 to version 4.2.4.

• Bug fixes:
  • Fixed the issue with the vulnerable 3rd Party component protobuf.
  • Fixed the issue with the vulnerable 3rd Party component openssl.
  • Fixed the issue with the wrong resource count for a particular project.
  • Fixed the issue with MLW RBAC not working if “user agent validation required” is checked in the Cumulocity IoT authentication settings (OAuth).