Welcome to the MLOPS wiki!

CI/CD (deployment centric) vs. ML Pipeline (experiment centric) views:

Feature Store idea:

Experiments idea:

Links:

ML/AI capabilities:

CACE: “Changing Anything Changes Everything”

Now see the animation in this page:

https://read.deeplearning.ai/the-batch/issue-91/

Machine Learning Development Lifecycle (MLDC)

https://github.com/aws-samples/mlops-amazon-sagemaker/blob/master/2-Bring-Your-Own/README.md

How different roles and tools interoperate in different MLOps maturity levels:

https://github.com/aws-samples/mlops-amazon-sagemaker/blob/master/2-Bring-Your-Own/README.md

Manual:

Repeatable:

Reliable:

Optimized:

Automation and Orchestration:

Apache Airflow:

KubeFlow:

KubeFlow Pipeline Components and Experiments:

Kubeflow and SageMaker Integration:

Introducing Amazon SageMaker Components for Kubeflow Pipelines

SageMaker Components for Kubeflow Pipelines

MLflow Components:

ML Platform:

MLOps can have many environments, pipelines and repositories:

MLOps Workload Orchestrator:

https://aws.amazon.com/solutions/implementations/mlops-workload-orchestrator/

Repositories:

SageMaker Projects:

ML Model Formats:

Packaging the model:

Deployment Patterns:

Inference infrastructures:

Real time inference:

Deployment Strategies

Standard Deployment:

B/G Deployments:

Shadow Deployment:

Deploy shadow ML models in Amazon SageMaker

Canary Deployment:

A/B Testing:

NOTE: You can use production variant for A/B testing and Canary Testing

Monitoring: