We've been following the latest Microsoft webinar about deploying the ML model from Azure Databricks to Azure ML using the MLFlow, and we get the following error when trying to run the experiment from Databricks notebook using the following code:
experimentName="someExperimentName"
mlflow.set_experiment(experimentName)
the error message:
UnsupportedModelRegistryStoreURIException: Unsupported URI 'adbazureml://westus.experiments.azureml.net/history/v1.0/subscriptions/cemrecdsap-t10us-20180830/resourceGroups/2f5a718e-7c56-4dd3-aa7b-03a19b70667/providers/Microsoft.MachineLearningServices/workspaces/cemrecdsap-mlservice' for model registry store. Supported schemes are: ['', 'file', 'sqlite', 'https', 'databricks', 'postgresql', 'mysql', 'http', 'mssql']
Init script we use as suggested in Microsoft MLflow webinar: (it was available here, but now it's removed - https://github.com/Azure/MachineLearningNotebooks/blob/master/how-to-use-azureml/azure-databricks/linking/azureml-cluster-init.sh)
#!/bin/bash
############## START CONFIGURATION #################
# Provide the required *AzureML* workspace information
region="westus"
subscriptionId="bcb65f42-f234-4bff-91cf-9ef816cd9936"
resourceGroupName="dev-rg"
workspaceName="myazuremlws"
# Optional config directory
configLocation="/databricks/config.json"
############### END CONFIGURATION #################
# Drop the workspace configuration on the cluster
sudo touch $configLocation
sudo echo {\\"subscription_id\\": \\"${subscriptionId}\\", \\"resource_group\\": \\"${resourceGroupName}\\", \\"workspace_name\\": \\"${workspaceName}\\"} > $configLocation
# Set the MLflow Tracking URI
trackingUri="adbazureml://${region}.experiments.azureml.net/history/v1.0/subscriptions/${subscriptionId}/resourceGroups/${resourceGroupName}/providers/Microsoft.MachineLearningServices/workspaces/${workspaceName}"
sudo echo export MLFLOW_TRACKING_URI=${trackingUri} >> /databricks/spark/conf/spark-env.sh
We use the latest MLFlow version, 1.4
Is there a chance that the adbazureml protocol that was used in the webinar is not supported yet by MLFlow? Or we missed something else?
