Known issues and limitations

Cloudera AI has the following known issues and limitations with experiments and MLflow.

• Cloudera AI currently supports only Python for experiment tracking.
• Experiment runs cannot be created from MLFlow on sessions using Legacy Engine. Instead, create a session using an ML Runtimes.
• The version column in the runs table is empty for every run. In a future release, this will show a git commit sha for projects using git.
• There is currently no mechanism for registering a model to a AI Registry. In a future release, you will be able to register models to a AI Registry and then deploy Model REST APIs with those models.
• Browsing an empty experiment will display a spinner that does not go away.
• Running an experiment from the workbench (from the dropdown menu) refers to legacy experiments and should not be used going forward.
• Tag/Metrics/Parameter columns that were previously hidden on the runs table will be remembered, but Cloudera AI won’t remember hiding any of the other columns (date, version, user, etc.)
• Admins can not browse all experiments. They can only see their experiments on the global Experiment page.
• Performance issues may arise when browsing the run details of a run with a lot of metric results, or when comparing a lot of runs.
• Runs can not be deleted or archived.