Summary
In the style of Andrew Ng's book Machine Learning Yearning, this workshop endeavors to introduce participants to the reasoning and architecture surrounding the productionization of a machine learning model. It will NOT delve into the specific architectural details of various types of machine learning models themselves (though I'll be prepared to answer questions, within a time box, on this) - the point is to understand:
- What does the loop look like of running, evaluating, and changing a machine learning model, and what technical details does the API of this workflow abstract?
- How might we put a model in production, and how will users access it?
- How can we swap out this model behind the API?
- How might we hide the specific language of ML behind good interfaces for consumers?
- What monitoring and testing should we implement during development, deployment, and running in production?
- What role does our context play in our choices about monitoring and error analysis?
For who?
This workshop targets professional software engineers with a few years of experience.
Learning Objectives
Participants will come away having done their own error analysis and made rudimentary changes to a model, then put that model in production in a manner very similar to that which might be used on the job. They'll have a firsthand experience with useful software engineering patterns more sophisticated than "fling model file on S3 and open it in a flask app," and they'll be able to speak to the engineering challenges specific to working with machine learning models.
Testimonial
Chelsea is one of the most engaging programming teachers I've learned from in recent memory,