Machine Learning

Release 10.13.0

Improvements in Machine Learning Engine

• Enhancements:
  • New API in Nyoka microservice to generate matrix profile for univariate time series data for detecting trends and anomalies.
  • Enabled Zementis microservice to scale out if and when more compute is required for serving incoming inference requests.
  • Implemented caching of user roles in Nyoka and Onnx microservices to improve performance.
  • Improved pipeline restoration in the Onnx microservice which ensures a more robust & reliable restoration process.
  • Added support for audio processing by allowing pre-processing scripts to leverage the librosa library in the Onnx microservice.
  • In the Machine Learning web application, ongoing requests are now canceled if the user navigates away from the current page at any point in time.

• Upgrades:
  • Machine Learning web application upgraded from Angular version 11 to version 12.
  • Machine Learning web application upgraded to use Cumulocity Web SDK 1013.0.0.
  • Nyoka and Onnx microservices upgraded NumPy from version 1.18.1 to version 1.18.5.
  • Zementis microservice upgraded to use Cumulocity Java Microservice SDK 1011.0.16.

• Bug fixes:
  • Fixed the issue with Access is denied when OAuth is used as the tenant’s authentication scheme.
  • Fixed the issue with Resource not found error message when trying to delete artifacts.
  • Fixed the issue with the artifact download when the artifact name contains a comma.
  • Fixed the issue with predictions when input data is provided in a ZIP format.

Improvements in Machine Learning Workbench

• Role Based Access Control:
  With this release, to use Cumulocity IoT Machine Learning Workbench, it is mandatory that users are assigned to Machine Learning global roles or user groups. This was done to tighten the security of the product and avoid issues related to privilege escalation. Existing Machine Learning Workbench application users will see an Access denied error if they are not yet part of Machine Learning global roles or user groups. Likewise, the MLW microservice APIs will respond with 403 - Forbidden if the necessary permissions are not associated with the requests made by the clients. Refer to the Machine Learning guide for more details.

• Enable Project Export and Import:
  Machine Learning Workbench provides a project-based structure for encapsulating data science assets - data, code, resources, and models. This helps data scientists and machine learning practitioners organize assets relevant to various stages of model lifecycle from data preparation to model building and deployment. To facilitate collaboration and sharing, users can now export the contents of a project as a compressed file and similarly, contents from an exported archive can be uploaded as new project.

• Notebook UI:
  The Notebook UI of Machine Learning Workbench was a custom implementation created by using the Monaco editor. The UI has been upgraded to use off-the-shelf JupyterLab UI components, which is the next-generation user interface for the project Jupyter, offering all the familiar building blocks of the classic Jupyter Notebook (notebook, text editor, rich outputs, etc.) in a flexible and powerful user interface.

• Enhancements:
  • Project improvements:
    • Enabled resource rename for project artifacts.
    • Enabled data validation with type annotations using the Pydantic package for all the MLW APIs.
    • Ability to delete a specific version of a project via the API ({{url}}/service/mlw/projects/{projectID}?versionNumber={versionNumber}).
    • Neural networks improvements:
    • Full screen mode for Neural Network Designer.
    • Improvements to Neural Network task to show model training progress graphically.

• Upgrades:
  • Machine Learning Workbench web application upgraded from Angular version 11 to version 12.
  • MLW microservice upgraded from tensorflow version 1.15.3 to version 2.0.4.
  • MLW microservice upgraded from ipython version 7.16.1 to version 7.31.1.
  • MLW microservice upgraded from ipykernel version 5.3.4 to version 6.8.0.
  • MLW microservice upgraded from Apache-httpd version 2.4.38 to version 2.4.52.
  • MLW microservice downgraded from numpy version 1.19.1 to version 1.18.5 to support tensorflow 2.0.

• Bug fixes:
  • Fixed the issue with downloading artifacts from a project taking long time by introducing the progress bar.
  • Fixed the issue with training more epochs on the integrated Jupyter Notebook within the MLW by upgrading tensorflow to 2.0.
  • Fixed the issue with committing the project when an existing project commit task is in-progress.
  • Fixed the issue with deleting the project when any of the project related task is in-progress.