A machine learning problem solving checklist

I use checklists to systematically approach problems, especially in domains where I have less experience. I apply each element in a checklist to reduce assumptions and avoid making common mistakes.

Here is my machine learning problem solving checklist:

• Understand the data structure from the first principles.

• Explicitly define the inputs and outputs of the system.

• Try to map the current problem to established methods before inventing anything new.

• Find out how other people have solved similar problems.

• Explicitly define evaluation metrics.

• Fit many models and empirically see which one is best on the evaluation metrics.

• Create an end-to-end system as quickly as possible.