Programming Study日本語

Disclosure: We earn a commission from Udemy when you buy a course through links on this page.

ML Model Deployment: Flask, FastAPI, Docker, Kubernetes

ML Model Deployment: Flask, FastAPI, Docker, Kubernetes

Udemy · English

About this course

Language
English
Category
Provider
Udemy
Data updated

You can fit a model. You have never been on call. That is the gap this course is for. Almost every deployment tutorial ends at the same place: the endpoint returns a prediction, the terminal shows some JSON, and the lesson stops. Everything expensive happens after that moment. This course keeps going. What the service does when real demand arrives. What OOMKilled means, why your pod died at three in the morning with no application log, and where the evidence actually is. Why your image rebuilds from scratch every single time because of one misplaced line in a Dockerfile. What a cold start costs you and how to live with it. And what you do in the twenty minutes after you realise the new version is worse than the old one. You deploy one model the whole way through: a small-business loan approval model, 12 features, trained on a seeded synthetic extract of 24,000 rows. It is invented, and so is the lender - no real organisation appears here as a case study. It scores 75.03% on held-out data against a 73.0% approval rate, which makes it a weak model, deliberately. A weak model deploys exactly like a strong one, and the deployment is the subject. A megabyte and a half of training data collapses into a 1,440-byte pickle, or 1,225 bytes as joblib, 14.93% smaller. The whole course is about moving that artefact around correctly. It teaches Flask and FastAPI, in that order, and takes a position on both. You build the Flask service, put gunicorn and nginx in front of it, and find out what workers and threads actually do. Then WSGI versus ASGI as a mechanism rather than a slogan: what an event loop buys you, what transfers unchanged, the four things that change in the code, typed requests with Pydantic and the documentation you get free. And then the lecture most courses skip - scoring a model is CPU-bound, it never awaits, and writing async def around it does not make it faster. It can make it worse, because one blocking cal…

Courses in the same category