RWML #037: Build and deploy an ML REST API in 5 steps
Read on realworldml.xyz
Here is a hands-on FREE tutorial on MLOps that will help you grow as an ML engineer and go beyond notebooks.
Tutorial materials 📚🎬
All the code is open-source and available in this repository
➡️ Give it a star ⭐ on GitHub to support my workAll the video lessons are on my Youtube channel
➡️ Subscribe to my channel to be notified when the next lesson is out.
This is what you will learn
You will learn to
train an ML model prototype through careful experimentation, using CometML.
deploy the model as a REST API, with Cerebrium.
automate safe deployments, using GitHub actions and Comet ML Model Registry [COMING SOON]
Without further ado, let's get to work!
1. Set up your development environment
Learn to
quickly bootstrap a project structure
manage Python environments and dependencies with Python Poetry
organize your code for maximum productivity
2. Generate training data
In real-world projects there is no training dataset waiting for you.
You need to generate it, starting from raw data, and ending up with clean pairs of (features, target).
This is what you will learn in this lecture.
3. Create a baseline model (without ML!)
Before training any ML model you need to estimate a baseline performance, using a simple rule-based model.
Simple historical values, or moving averages often work well.
4. Train ML models
Building a good ML model is a very experimental process.
Integrate your training script with a serverless experiment tracker like CometML so you experiment faster in a more disciplined way.
5. Deploy your model as a Serverless REST API
Un-deployed models generate 0 business value.
Make your model predictions available by deploying the model as a Serverless REST API using Cerebrium
What’s coming next?
In the following weeks I will release a few more videos, where you will learn how to automate deployments following CI/CD best-practices.
Let’s keep on learning.
Peace and Love.
Pau