Learn MLOps with This Free Course
Learn to train and track your experiments, create ML pipelines, model deployment, monitor the performance in production, and adopt best practices from DevOps.
Image by Author | Canva Pro
Table of Contents
- What is MLOps?
- Why Do We Need MLOps?
- MLOps Zoomcamp
- Final Thoughts
- Frequently Asked Questions
What is MLOps?
MLOps stands for machine learning operations. The term MLOps is derived from DevOps (Development Operations). It is used to streamline the machine learning process from development to deployment. The MLOps include training machine learning models, experiment tracking, model optimization, creating ML pipelines, saving and serving models, and monitoring and maintaining models in production.
In short, you are automating all the processes from development to deployment, and you are constantly monitoring the logs, metrics, and performance.
Why Do We Need MLOps?
Deploying machine learning models in production is complex and challenging. A typical machine learning life cycle consists of data ingestion, data augmentation, modeling, training, optimizing, tracking experiments, and explaining the results. It requires the collaboration of an entire data team, from data engineers to data scientists. MLOps makes these processes automated so that you can continuously monitor and improve the model in production.
MLOps allows you to reduce model deployment time and deliver higher-quality ML models. It provides vast scalability and management of thousands of ML models. You can control, manage, and monitor the continuous integration, continuous delivery, and continuous deployment - Databricks.
MLOps Zoomcamp teaches you practical aspects of model deployment and monitoring. This course is not for beginners and requires you to know Python, Docker, command line, a deep understanding of machine learning models, and at least 1 year of programming experience. If you are a complete beginner I will suggest you take free Machine Learning Zoomcamp before you start learning MLOps.
The course consists of seven modules and a project. You can sign up for the course here.
Module 1: Introduction
We will learn the basics of MLOps, course overview, and model maturity. We will also learn to load a parquet file of the NY Taxi trips dataset and train a ride duration model.
Module 2: Experiment tracking
We will learn about MLflow and the best practices of experiment tracking. We will also learn to save and load models using MLflow and learn about the model registry.
Module 3: Orchestration and ML Pipelines
We will learn about machine learning pipelines and use Prefect to orchestrate machine learning projects. We will be turning Jupyter notebooks into ML pipelines and learning about the popular tool KubeFlow pipelines.
Module 4: Model Deployment
We will learn about batch vs online deployment and web services versus streaming. We will serve models in batch mode, learn about web services, and create streaming using Kinesis/SQS + AWS Lambda.
Module 5: Model Monitoring
We will learn the difference between ML monitoring and software monitoring, data quality monitoring, data drift and concept drift, and batch versus real-time monitoring. We will learn to use Evidently, Prometheus, and Grafana to monitor models in production.
Module 6: Best Practices
We will adopt best practices from DevOps in machine learning workflow:
- Virtual environments and Docker
- Python logging and linting
- Testing unit, integration, and regression
- CI/CD (github actions)
- Infrastructure as code (Terraform, cloud formation)
Module 7: Processes
We will learn about the MLOps process and planning using CRISP-DM, CRISP-ML, ML Canvas, Data Landscape canvas, and Documentation practices in ML projects (Model Cards Toolkit.
Finally, we will incorporate the tools and practices we have learned into one end-to-end project. We are going to use the NY Taxi trips dataset to build an MLOps system to predict the ride duration or if the driver is going to be tipped or not.
Every week you will join a live session on the DataTalksClub YouTube channel and learn about each module. You can also check the complete playlist here. At the end of each module, you will be given homework, so that you can retain the knowledge for the next module after a week.
I am a big fan of DataTalks.Club free courses, as they provide practical knowledge on machine learning, data engineering, and MLOps. If you are searching for the best free data engineering or MLOps courses, you will find none. DataTalks.Club is democratizing machine learning so that everyone can learn machine learning operations for free.
If you are a machine learning enthusiast, try to learn MLOps to excel in your career and understand the data ecosystem. From data ingestions to production.
You can also check out courses from DataTalks.Club:
- Machine Learning Zoomcamp - free 4-month course about ML Engineering
- Data Engineering Zoomcamp - free 9-week course about Data Engineering
Frequently Asked Questions
What are MLOps tools?
- For tracking ML experiments, there is MLflow and Comet.
- For Data and Model versioning, there is DVC, DAGsHub, and Pachyderm.
- For optimizing experiments, there is Optuna and Sigopt.
- For workflow pipelines and Orchestration, there is Kubeflow and Apache Airflow.
- For model saving and serving, there is BentoML and Cortex.
- For monitoring the models in production, there is Evidently and NewRelic.
How is MLOps different from DevOps?
- DevOps is a set of practices to improve the software development process by continuous integration and delivery of high-quality development products.
- MLOps is the process of automating the deployment of machine learning applications and monitoring the performance in production.
Why MLOps is Important?
Deploying machine learning models is costly, and they don’t generate value to sustain the business. MLOps allows companies to easily deploy, monitor, and update models in production.
What is the best free MLOps course?
MLOps Zoomcamp is the best and free for everyone. Other than that, you won’t find free complete MLOps courses on the web.
How to get MLOps certifications?
Currently, there are no certifications on MLOps. Most organizations will accept AWS Certified DevOps Engineer - Professional Certification for the MLOps engineering position.
Abid Ali Awan (@1abidaliawan) is a certified data scientist professional who loves building machine learning models. Currently, he is focusing on content creation and writing technical blogs on machine learning and data science technologies. Abid holds a Master's degree in Technology Management and a bachelor's degree in Telecommunication Engineering. His vision is to build an AI product using a graph neural network for students struggling with mental illness.