Connect on-prem jypyter notebook to mlflow tracking server in Azure

Question

Is it possible to connect a notebook running in premises to an mlflow Tracking server that is part of an Azure Databricks workspace? Have all the local logging and tracking saved in Azure?

Answer 1

I had a similar problem, used python and solved it with the following steps:

1. Install mlflow and datbricks-cli libraries.
2. Define the following env variables : DATABRICKS_HOST (databricks workspace url: https://region.azuredatabricks.net) and DATABRICKS_TOKEN
3. Define mlflow client:

mlflow_client = mlflow.tracking.MlflowClient(tracking_uri='databricks')

5. Use mlflow_client client for logging, saving and etc..

for more reference you can look at the "Log to a tracking server from a notebook" section here

Answer 2

Following the steps in the accepted answer, I couldn't use the mlflow_client object. What worked is setting the tracking uri directly on mlflow:

mlflow.set_tracking_uri('databricks')