datarobot-mlflow 0.1.dev2

datarobot-mlflow client to synchronize an MLFlow model with DataRobot model

mlflow-integration

Provide means of exporting a model from MLFlow model registry and pushing it to DataRobot model registry

Key-values are created from training parameters, metrics, tags, and artifacts in the MLflow model.

Setup

• Python 3.7 or later
• DataRobot 9.0 or later
• pip install datarobot-mlflow
  • if using Azure: pip install "datarobot-mlflow[azure]"

Considerations

This integration library uses an API endpoint under Public Preview. The DataRobot user owning the API token used below must have:

• Enable Extended Compliance Documentation set
• Owner or User permission for the DataRobot model package

DataRobot information needed

• URL of DataRobot instance, example: https://app.datarobot.com
• ID of the model package to receive key-values; example: 64227b4bf82db411c90c3209
• API token for DataRobot: export MLOPS_API_TOKEN=<API token from DataRobot Developer Tools>

Local MLflow information needed

• MLflow tracking URI; example "file:///Users/me/mlflow/examples/mlruns"
• Model name; example "cost-model"
• Model version; example "2"

Azure DataBricks MLFlow with Service Principal information needed

• MLflow tracking URI; example "azureml://region.api.azureml.ms/mlflow/v1.0/subscriptions/subscription-id/resourceGroups/resource-group-name/providers/Microsoft.MachineLearningServices/workspaces/azure-ml-workspace-name"
• Model name; example "cost-model"
• Model version; example "2"
• Provide service principal details in environment:
  • export AZURE_TENANT_ID="<tenant-id>"
  • export AZURE_CLIENT_ID="<client-id>"
  • export AZURE_CLIENT_SECRET="<secret>"

Example: Import from MLflow

DR_MODEL_ID="<MODEL_PACKAGE_ID>"

env PYTHONPATH=./ \
python datarobot_mlflow/drflow_cli.py \
  --mlflow-url http://localhost:8080 \
  --mlflow-model cost-model  \
  --mlflow-model-version 2 \
  --dr-model $DR_MODEL_ID \
  --dr-url https://app.datarobot.com \
  --with-artifacts \
  --verbose \
  --action sync

Example: validate Azure credentials

export MLOPS_API_TOKEN="n/a"  # not used for Azure auth check, but must be present

env PYTHONPATH=./ \
python datarobot_mlflow/drflow_cli.py \
  --verbose \
  --auth-type azure-service-principal \
  --service-provider-type azure-databricks \
  --action validate-auth

# example output for missing environment variables:
Required environment variable is not defined: AZURE_TENANT_ID
Required environment variable is not defined: AZURE_CLIENT_ID
Required environment variable is not defined: AZURE_CLIENT_SECRET
Azure AD Service Principal credentials are not valid; check environment variables

# example output for successful authentication:
Azure AD Service Principal credentials are valid for obtaining access token

Actions

The following operations are available for --action:

• sync: import parameters, tags, metrics, and artifacts from MLflow model.
• list-mlflow-keys: list parameters, tags, metrics, and artifacts in an MLflow model. Requires --mlflow-url, --mlflow-model, and --mlflow-model-version.
• validate-auth: see "validate Azure credentials" example above.

Options

The following options can be added to the drflow_cli command line:

• --mlflow-url: MLflow Tracking URI
• --mlflow-model: MLflow model name
• --mlflow-model-version: MLflow model version
• --dr-url: Main URL of the DataRobot instance
• --dr-model: DataRobot Model Package ID. Registered Model Versions are also supported.
• --prefix: a string to prepend to the names of all key-values posted to DataRobot. Default is empty.
• --debug: set Python logging level to logging.DEBUG. Default level is logging.WARNING.
• --verbose: prints to stdout information about the following:
  • retrieving model from MLflow; prints model information
  • setting model data in DataRobot: prints each key-value posted
• --with-artifacts: download MLflow model artifacts to /tmp/model
• --service-provider-type: service provider to use for validate-auth. Supported values are:
  • azure-databricks: for Databricks MLflow within Azure
• --auth-type: authentication type for validate-auth. Supported values are:
  • azure-service-principal: for Azure Service Principal