Explore the unique challenges and solutions in machine learning programming through the lens of a software engineer in this conference talk from code::dive 2018. Discover the differences between traditional software engineering and machine learning practices, including the focus on numerical experiments, changing requirements, and the potential for misleading outputs. Learn about the culture of machine learning programming, unexpected challenges, and problem-solving approaches specific to this rapidly evolving field. Gain insights into biomarker identification, decision trees, prototype testing, data collection, model components, and the complexities of choosing algorithms. Understand the importance of generalization, handling unseen data, and evaluating model performance in the context of machine learning development.
Overview
Syllabus
Intro
"Starter" problem: biomarkers identification
Let's start from something simpler
Decision trees: Shape.
Decision trees What questions?
I've got a prototype! Testing?
Original data collection
The Pipeline: summary
Two components of a good model
The Black Box
What could go wrong?
Generalization
Unseen data
Which algorithm is better?
So how do I choose?
This sounds like a lot of work!
Taught by
code::dive conference
