Track ML models with MLflow and Azure Machine Learning

In this article, learn how to enable MLflow's tracking URI and logging API, collectively known as MLflow Tracking, to connect Azure Machine Learning as the backend of your MLflow experiments.

Supported capabilities include:

MLflow is an open-source library for managing the life cycle of your machine learning experiments. MLflow Tracking is a component of MLflow that logs and tracks your training run metrics and model artifacts, no matter your experiment's environment--locally on your computer, on a remote compute target, a virtual machine, or an Azure Databricks cluster.

See MLflow and Azure Machine Learning for additional MLflow and Azure Machine Learning functionality integrations.

The following diagram illustrates that with MLflow Tracking, you track an experiment's run metrics and store model artifacts in your Azure Machine Learning workspace.

mlflow with azure machine learning diagram

Tip

The information in this document is primarily for data scientists and developers who want to monitor the model training process. If you are an administrator interested in monitoring resource usage and events from Azure Machine Learning, such as quotas, completed training runs, or completed model deployments, see Monitoring Azure Machine Learning.

Note

You can use the MLflow Skinny client which is a lightweight MLflow package without SQL storage, server, UI, or data science dependencies. This is recommended for users who primarily need the tracking and logging capabilities without importing the full suite of MLflow features including deployments.

Prerequisites

Track local runs

MLflow Tracking with Azure Machine Learning lets you store the logged metrics and artifacts runs that were executed on your local machine into your Azure Machine Learning workspace.

Set up tracking environment

To track a local run, you need to point your local machine to the Azure Machine Learning MLflow Tracking URI.

Import the mlflow and Workspace classes to access MLflow's tracking URI and configure your workspace.

In the following code, the get_mlflow_tracking_uri() method assigns a unique tracking URI address to the workspace, ws, and set_tracking_uri() points the MLflow tracking URI to that address.

import mlflow
from azureml.core import Workspace

ws = Workspace.from_config()

mlflow.set_tracking_uri(ws.get_mlflow_tracking_uri())

Set experiment name

All MLflow runs are logged to the active experiment, which can be set with the MLflow SDK or Azure CLI.

Set the MLflow experiment name with set_experiment() command.

experiment_name = 'experiment_with_mlflow'
mlflow.set_experiment(experiment_name)

Start training run

After you set the MLflow experiment name, you can start your training run with start_run(). Then use log_metric() to activate the MLflow logging API and begin logging your training run metrics.

import os
from random import random

with mlflow.start_run() as mlflow_run:
    mlflow.log_param("hello_param", "world")
    mlflow.log_metric("hello_metric", random())
    os.system(f"echo 'hello world' > helloworld.txt")
    mlflow.log_artifact("helloworld.txt")

Track remote runs

Remote runs let you train your models on more powerful computes, such as GPU enabled virtual machines, or Machine Learning Compute clusters. See Use compute targets for model training to learn about different compute options.

MLflow Tracking with Azure Machine Learning lets you store the logged metrics and artifacts from your remote runs into your Azure Machine Learning workspace. Any run with MLflow Tracking code in it will have metrics logged automatically to the workspace.

First, you should create a src subdirectory and create a file with your training code in a train.py file in the src subdirectory. All your training code will go into the src subdirectory, including train.py.

The training code is taken from this MLflow example in the Azure Machine Learning example repo.

Copy this code into the file:

# imports
import os
import mlflow

from random import random

# define functions
def main():
    mlflow.log_param("hello_param", "world")
    mlflow.log_metric("hello_metric", random())
    os.system(f"echo 'hello world' > helloworld.txt")
    mlflow.log_artifact("helloworld.txt")

# run functions
if __name__ == "__main__":
    # run main function
    main()

Load training script to submit an experiement.

script_dir = "src"
training_script = 'train.py'
with open("{}/{}".format(script_dir,training_script), 'r') as f:
    print(f.read())

In your script, configure your compute and training run environment with the Environment class.

from azureml.core import Environment
from azureml.core.conda_dependencies import CondaDependencies

env = Environment(name="mlflow-env")

# Specify conda dependencies with scikit-learn and temporary pointers to mlflow extensions
cd = CondaDependencies.create(
    conda_packages=["scikit-learn", "matplotlib"],
    pip_packages=["azureml-mlflow", "pandas", "numpy"]
    )

env.python.conda_dependencies = cd

Then, construct ScriptRunConfig with your remote compute as the compute target.

from azureml.core import ScriptRunConfig

src = ScriptRunConfig(source_directory="src",
                      script=training_script,
                      compute_target="<COMPUTE_NAME>",
                      environment=env)

With this compute and training run configuration, use the Experiment.submit() method to submit a run. This method automatically sets the MLflow tracking URI and directs the logging from MLflow to your Workspace.

from azureml.core import Experiment
from azureml.core import Workspace
ws = Workspace.from_config()

experiment_name = "experiment_with_mlflow"
exp = Experiment(workspace=ws, name=experiment_name)

run = exp.submit(src)

View metrics and artifacts in your workspace

The metrics and artifacts from MLflow logging are tracked in your workspace. To view them anytime, navigate to your workspace and find the experiment by name in your workspace in Azure Machine Learning studio. Or run the below code.

Retrieve run metric using MLflow get_run().

from mlflow.entities import ViewType
from mlflow.tracking import MlflowClient

# Retrieve run ID for the last run experiement
current_experiment=mlflow.get_experiment_by_name(experiment_name)
runs = mlflow.search_runs(experiment_ids=current_experiment.experiment_id, run_view_type=ViewType.ALL)
run_id = runs.tail(1)["run_id"].tolist()[0]

# Use MlFlow to retrieve the run that was just completed
client = MlflowClient()
finished_mlflow_run = MlflowClient().get_run(run_id)

metrics = finished_mlflow_run.data.metrics
tags = finished_mlflow_run.data.tags
params = finished_mlflow_run.data.params

print(metrics,tags,params)

Retrieve artifacts with MLFLow

To view the artifacts of a run, you can use MlFlowClient.list_artifacts()

client.list_artifacts(run_id)

To download an artifact to the current directory, you can use MLFlowClient.download_artifacts()

client.download_artifacts(run_id, "helloworld.txt", ".")

Compare and query

Compare and query all MLflow runs in your Azure Machine Learning workspace with the following code. Learn more about how to query runs with MLflow.

from mlflow.entities import ViewType

all_experiments = [exp.experiment_id for exp in MlflowClient().list_experiments()]
query = "metrics.hello_metric > 0"
runs = mlflow.search_runs(experiment_ids=all_experiments, filter_string=query, run_view_type=ViewType.ALL)

runs.head(10)

Automatic logging

With Azure Machine Learning and MLFlow, users can log metrics, model parameters and model artifacts automatically when training a model. A variety of popular machine learning libraries are supported.

To enable automatic logging insert the following code before your training code:

mlflow.autolog()

Learn more about Automatic logging with MLflow.

Manage models

Register and track your models with the Azure Machine Learning model registry, which supports the MLflow model registry. Azure Machine Learning models are aligned with the MLflow model schema making it easy to export and import these models across different workflows. The MLflow related metadata such as, run ID is also tagged with the registered model for traceability. Users can submit training runs, register, and deploy models produced from MLflow runs.

If you want to deploy and register your production ready model in one step, see Deploy and register MLflow models.

To register and view a model from a run, use the following steps:

  1. Once a run is complete, call the register_model() method.

    # the model folder produced from a run is registered. This includes the MLmodel file, model.pkl and the conda.yaml.
    model_path = "model"
    model_uri = 'runs:/{}/{}'.format(run_id, model_path) 
    mlflow.register_model(model_uri,"registered_model_name")
    
  2. View the registered model in your workspace with Azure Machine Learning studio.

    In the following example the registered model, my-model has MLflow tracking metadata tagged.

    register-mlflow-model

  3. Select the Artifacts tab to see all the model files that align with the MLflow model schema (conda.yaml, MLmodel, model.pkl).

    model-schema

  4. Select MLmodel to see the MLmodel file generated by the run.

    MLmodel-schema

Clean up resources

If you don't plan to use the logged metrics and artifacts in your workspace, the ability to delete them individually is currently unavailable. Instead, delete the resource group that contains the storage account and workspace, so you don't incur any charges:

  1. In the Azure portal, select Resource groups on the far left.

    Delete in the Azure portal

  2. From the list, select the resource group you created.

  3. Select Delete resource group.

  4. Enter the resource group name. Then select Delete.

Example notebooks

The MLflow with Azure ML notebooks demonstrate and expand upon concepts presented in this article. Also see the community-driven repository, AzureML-Examples.

Next steps