MLflow is a comprehensive Machine Learning Operations (MLOps) service that facilitates the monitoring, maintenance, and continuous integration and deployment of Machine Learning models.

The MLflow Agent allows you to choose a published linear regression model from the MLflow repository, invoke the model with input value(s), and obtain the predicted outcome from the formula.

Or choose a staging model from the MLflow repository to verify the accuracy level before it is transitioned to Production.

Or choose a specific model version (that is not yet in Staging or Production) to run live data against the model in a Data Stream.

The Agent enables effective governance, empowering data scientists to promote a new version of the model within MLflow without necessitating any edits or republishing of the XMPro Data Stream. This promotion takes effect as long as long as the new version retains identical input/output properties.

Details for an example and its configuration can be found in the How to Use section.

More information about the MLflow service can be found here.


Access to the following is required to use this Agent:

  • An MLflow server, which serves details about experiments and published models.

  • A signed, published MLflow model, that receives values and returns a prediction.

    • A signature is logged with the MLflow model

    • The standard data types supported for input and output are Double, Float, Integer, Boolean, DateTime, Long, and String. The Tensor data formats are not supported.

At the time of writing, MLflow servers do not offer inbuilt methods of authentication. This Agent assumes that access to the infrastructure will not require authentication.

Current Version

Please contact XMPro if you're looking for an older version of this Agent.

MLflow Release Notes

v1.06, 2 Apr 2024

  • Moved polling property to the Model group

v1.05, 14 Mar 2024

  • Support Staging models (refer the new Model Usage property)

  • Support a specific model version (refer the new Model Usage property)

  • Do not poll if a specific version has been configured

  • Validate that the model version can be found i.e. the production version or staging version or the specified version

  • Validation that there is a model signature

v1.04, 20 Jul 2023

  • Fixed the blank Model property if the version at the time of configuration is archived. The Model property automatically selects the most recently published version of the configured model, matching the output payload's ModelVersion attribute behavior.

v1.03, 27 Jun 2023

  • Initial release

Last updated