Options
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
01-03-2023 07:14 AM
We are using the managed mlflow, but we want to access the metadata of the models and show it in another application. There is already a server that I can query?
Can I re-create/configure the databricks workspace to make the managed mlflow use a postgresql server?
Options
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
01-03-2023 07:40 AM
Ideas which I have is:
- periodically export/import mlflow models and experiments https://github.com/mlflow/mlflow-export-import#why-use-mlflow-export-import
- get metadata through API https://docs.databricks.com/dev-tools/api/latest/mlflow.html#operation/get-registered-model
- when you run your experiments in Databricks in notebooks, you can change the tracking server,
- I haven't heard about the availability to change the database server for registered models and experiments in managed Databricks Mlflow (although you can ask your databricks representative),
- It can change in the future. Recently it was announced that it is possible to use external feature stores https://docs.databricks.com/machine-learning/feature-store/online-feature-stores.html, so it could be nice to have similar options with models and experiments.
My blog: https://databrickster.medium.com/
Options
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
01-03-2023 09:21 AM
Thanks for the complete answer. It was very helpful.
