Azure Machine Learning

Overview

Deploying a trained model to Azure Machine Learning enables you to expose it as a real‑time endpoint or a batch scoring pipeline. This guide walks you through the supported deployment options, best practices, and how to monitor your service once it’s live.

Prerequisites

• An Azure subscription with access to Azure Machine Learning workspace.
• Azure CLI 2.30+ or az ml extension installed.
• Python 3.8‑3.11 with azureml-core and azureml-mlflow packages.
• Trained model artifact (e.g., .pkl, .onnx, or .tar.gz).

Deploy with Azure ML SDK

from azureml.core import Workspace, Model, Environment, InferenceConfig
from azureml.core.webservice import AciWebservice, Webservice

ws = Workspace.from_config()
model = Model.register(workspace=ws,
                       model_path="outputs/model.pkl",
                       model_name="my-model")

env = Environment.from_conda_specification(
    name="my-env",
    file_path="environment.yml"
)

inference_config = InferenceConfig(entry_script="score.py", environment=env)

deployment_config = AciWebservice.deploy_configuration(cpu_cores=2,
                                                      memory_gb=4,
                                                      auth_enabled=True)

service = Model.deploy(workspace=ws,
                       name="my-model-service",
                       models=[model],
                       inference_config=inference_config,
                       deployment_config=deployment_config)
service.wait_for_deployment(show_output=True)

print(f"Scoring URI: {service.scoring_uri}")
print(f"Swagger URI: {service.swagger_uri}")

# Cell 1: Load workspace
ws = Workspace.from_config()

# Cell 2: Register model
model = Model.register(ws, model_path="model.pkl", model_name="my-model")

# Cell 3: Define env & inference
env = Environment.from_conda_specification("ml-env", "environment.yml")
inf_cfg = InferenceConfig(entry_script="score.py", environment=env)

# Cell 4: Deploy to ACI
aci_cfg = AciWebservice.deploy_configuration(cpu_cores=2, memory_gb=4)
svc = Model.deploy(ws, "my-model-svc", [model], inf_cfg, aci_cfg)
svc.wait_for_deployment(show_output=True)
svc.scoring_uri

Deploy with Azure CLI

Run the following commands from your terminal. Make sure you have az ml extension installed.

# Register the model
az ml model register -n my-model -f model.pkl

# Create an environment definition (environment.yml)
az ml environment create -f environment.yml -n my-env

# Deploy the model to Azure Container Instance
az ml online-endpoint create -n my-model-endpoint --auth-mode token
az ml online-deployment create -e my-model-endpoint -n default \
    --model my-model:1 \
    --environment my-env:latest \
    --instance-type Standard_DS2_v2 \
    --scale-minimum 1 --scale-maximum 5

Monitoring & Scaling

Azure Machine Learning provides built-in metrics and autoscaling. Use the portal or CLI to configure alerts.

• Metrics: Requests, Latency, CPU, Memory.
• Autoscale: Based on CPU or Requests thresholds.
• Logging: Enable Application Insights for detailed request logs.

Example CLI to enable autoscaling:

az ml online-deployment update -e my-model-endpoint -n default \
    --scale-minimum 2 --scale-maximum 10 \
    --scale-rule-name cpu-rule \
    --scale-rule-type cpu \
    --scale-rule-threshold 70

Sample Code

Download a complete example project that includes score.py, environment.yml, and a Jupyter notebook.

FAQ