MEPX
Chapter 1 of 10All chapters
  1. 1Framing the problem
  2. 2Data
  3. 3Training and testing
  4. 4Underfitting and overfitting
  5. 5Measuring performance
  6. 6Common algorithms
  7. 7How training works
  8. 8The pipeline
  9. 9After deployment
  10. 10Common pitfalls

Chapter 1 of 10

Framing the problem

Before any code.

Is it a learning problem

Machine learning suits problems with a pattern, plenty of examples, and tolerance for being sometimes wrong. Rules are better when the logic is known and errors are unacceptable.

  • Write down what a correct prediction looks like and what an error costs.
  • If a simple rule reaches 90 percent, the model must beat that to be worth it.

Kinds of task

Classification predicts a category, regression predicts a number, clustering finds groups, and ranking orders items. The choice shapes the data, the model and the metric.