Transition from DevOps expert to MLOps expert

Introduction

Laymen explanation

MLOps, or machine learning operations, are the methods used to streamline the machine learning life cycle from beginning to end.

In loosely speaking, MLOps is nothing but a DevOps with additional ML addition in the workflow. There are things which needs to be added in your DevOps profile for tagging yourself as MLOps engineer. If you are interested to know them, then this document is for you.

Technical explanation

MLOps is an emerging practice of collaboration and communication between data engineers, data scientists, developers, and traditional DevOps to manage the ML lifecycle.

Relation between MLOps and DevOps

MLOps life cycle in view of DevOps

Example differences

• • MLOps expert understands that he needs to streamline/pipeline ML experimentation as well. For example, ML team wants to retrieve previously done model having specific parameters/hyperparameters set.

  • MLOps expert understands that data versioning is also important since the ML performance depends on training data

Helpful orchestration tools

Apache airflow, kubeflow, mlflow are few useful ML orchestrations tools for the same. Below is adoption comparative data below(Thanks to this article).

More about MLOps

Refer here for ML pipeline detail.

For ML lifecycle, refer this write-up

Reference

https://images.app.goo.gl/BWiGzBrK3CmzMBRj8

https://images.app.goo.gl/m4e6ynGraYXBbJn7A

https://images.app.goo.gl/WGLoRip45nKrXRit9

https://medium.com/@vsistla/why-mlops-shouldnt-be-an-afterthought-b73c564b96d7

https://valohai.com/blog/difference-between-devops-and-mlops/

https://www.linkedin.com/posts/dpkumar_pipeline-kubernetes-datascience-activity-6770150099663253504-niET

https://www.linkedin.com/posts/dpkumar_machinelearning-mlops-datascience-activity-6756195578104958976-0eOc

https://www.datarevenue.com/en-blog/airflow-vs-luigi-vs-argo-vs-mlflow-vs-kubeflow

https://www.linkedin.com/posts/dpkumar_mlops-devops-machinelearningtools-activity-6778307180220284928-gqvm