Cloud Deployment Platforms (AWS SageMaker / GCP Vertex)

Managed cloud ML platforms like AWS SageMaker and GCP Vertex AI abstract away infrastructure management, providing scalable training, one-click deployment, and built-in monitoring for production ML workloads.

AWS SageMaker

SageMaker offers a comprehensive suite covering data labelling, training jobs, hyperparameter tuning, real-time endpoints, batch transform, and model monitoring — all governed by IAM roles and VPC networking.

Deploying a Scikit-Learn Model to a SageMaker Endpoint

<pre><code class="language-python"># Using the SageMaker Python SDK import sagemaker from sagemaker.sklearn import SKLearnModel # Upload model artifact to S3 sagemaker_session = sagemaker.Session() model_s3_uri = sagemaker_session.upload_data( path="model.tar.gz", # tarball of model.joblib bucket="my-ml-bucket", key_prefix="models/iris-rf" ) # Define the model sklearn_model = SKLearnModel( model_data=model_s3_uri, role="arn:aws:iam::123456789012:role/SageMakerRole", entry_point="inference.py", # defines model_fn, predict_fn framework_version="1.2-1", py_version="py3" ) # Deploy as a real-time endpoint predictor = sklearn_model.deploy( initial_instance_count=1, instance_type="ml.t2.medium" ) # Invoke import numpy as np predictor.predict(np.array([[5.1, 3.5, 1.4, 0.2]]))</pre>

SageMaker Key Features

  • Training Jobs: Managed distributed training with spot instances
  • Autopilot: AutoML for tabular data
  • Model Monitor: Built-in data quality and drift monitoring
  • Pipelines: Directed acyclic graph (DAG) ML workflows
  • Feature Store: Centralised, versioned feature repository

GCP Vertex AI

Vertex AI unifies Google Cloud's ML services under a single API, offering custom training, AutoML, managed endpoints, Feature Store, Model Registry, and Vertex Pipelines (Kubeflow-based).

Deploying to a Vertex AI Endpoint

<pre><code class="language-python">from google.cloud import aiplatform aiplatform.init(project="my-gcp-project", location="us-central1") # Upload model (pre-built sklearn container) model = aiplatform.Model.upload( display_name="iris-random-forest", artifact_uri="gs://my-bucket/models/iris-rf/", serving_container_image_uri="us-docker.pkg.dev/vertex-ai/prediction/sklearn-cpu.1-4:latest" ) # Create endpoint and deploy endpoint = aiplatform.Endpoint.create(display_name="iris-endpoint") endpoint.deploy( model=model, machine_type="n1-standard-2", min_replica_count=1, max_replica_count=5 # auto-scales ) # Predict response = endpoint.predict(instances=[[5.1, 3.5, 1.4, 0.2]]) print(response.predictions)</pre>

Choosing Between Platforms

The right platform depends on your organisation's existing cloud footprint, data residency requirements, and preferred pricing model.

Quick Comparison

  • AWS SageMaker: Most mature, widest feature set, deep AWS ecosystem integration
  • GCP Vertex AI: Strong AutoML, native BigQuery/BigTable integration, Kubeflow pipelines
  • Azure ML: Best for Microsoft-centric shops, strong enterprise identity integration
  • All three support custom containers, so you're not locked into their native SDKs