Look Before You Model
What can I learn about the problem before choosing the solution?
Situation
You are tempted to begin with a technique, architecture or tool.
The idea
Understanding the raw problem can reveal structure, noise, constraints and opportunities that determine what kind of solution is appropriate.
Useful when
ML/data problemsproduct discoverytechnical designchoosing tools
Watch for
Inspection can itself become analysis paralysis. The objective is better hypotheses, not perfect understanding.
Where this came from
Applied machine learning