Exemplos práticos

Python — cria um job de treino em compute cluster.

from azure.ai.ml import MLClient, command, Input
from azure.identity import DefaultAzureCredential

ml = MLClient(DefaultAzureCredential(), "<sub>", "<rg>", "<workspace>")

job = command(
    code="./src",
    command="python train.py --data ${{inputs.data}} --epochs 10",
    inputs={"data": Input(type="uri_folder", path="azureml://datastores/main/paths/train/")},
    environment="azureml://registries/azureml/environments/sklearn-1.5/labels/latest",
    compute="cpu-cluster",
    display_name="train-churn-v1",
    experiment_name="churn",
)
returned_job = ml.jobs.create_or_update(job)
print(returned_job.studio_url)

az ml model create com path e tags.

az ml model create \
  --name churn-classifier \
  --version 3 \
  --path ./outputs/model \
  --type mlflow_model \
  --tags env=prod owner=data-team \
  --resource-group rg-mlops \
  --workspace-name ws-mlops

YAML de endpoint + deployment com traffic split.

# endpoint.yml
$schema: https://azuremlschemas.azureedge.net/latest/managedOnlineEndpoint.schema.json
name: churn-ep
auth_mode: key
---
# deployment.yml
$schema: https://azuremlschemas.azureedge.net/latest/managedOnlineDeployment.schema.json
name: blue
endpoint_name: churn-ep
model: azureml:churn-classifier:3
instance_type: Standard_DS3_v2
instance_count: 2
# az ml online-endpoint update -n churn-ep --traffic "blue=90 green=10"

curl com token Bearer e payload JSON.

TOKEN=$(az ml online-endpoint get-credentials -n churn-ep --query primaryKey -o tsv)

curl -X POST https://churn-ep.eastus.inference.ml.azure.com/score \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"input_data":{"columns":["age","balance"],"data":[[42,15000]]}}'

Stages: build, test, train, register, deploy.

# azure-pipelines.yml
trigger: [ main ]
pool: { vmImage: ubuntu-latest }

stages:
  - stage: Build
    jobs:
      - job: build
        steps:
          - script: pip install -r requirements.txt && pytest tests/
  - stage: Train
    jobs:
      - job: train
        steps:
          - task: AzureCLI@2
            inputs:
              azureSubscription: sc-azureml
              scriptType: bash
              inlineScript: az ml job create -f jobs/train.yml --stream
  - stage: Deploy
    dependsOn: Train
    jobs:
      - deployment: prod
        environment: production
        strategy:
          runOnce:
            deploy:
              steps:
                - script: az ml online-deployment create -f deploy.yml

Baseline vs. produção via SDK.

from azure.ai.ml.entities import (
    MonitorSchedule, MonitorDefinition, DataDriftSignal,
    BaselineDataRange, ProductionData, ReferenceData,
)

monitor = MonitorSchedule(
    name="churn-drift",
    create_monitor=MonitorDefinition(
        compute={"instance_type": "Standard_E4s_v3", "runtime_version": "3.3"},
        monitoring_signals={
            "data_drift": DataDriftSignal(
                production_data=ProductionData(input_data="azureml://endpoints/churn-ep/logs"),
                reference_data=ReferenceData(input_data="azureml:churn-baseline:1"),
                alert_enabled=True,
            ),
        },
    ),
)