Introduction
The whole point of machine learning is to do well on new examples, not just the ones it was trained on. This ability is called generalisation. There are two broad ways to achieve it.
Imagine you are guessing the price of a second-hand car:
- Approach A: you remember similar cars you've seen for sale — same model, similar age and mileage — and say "it's probably about what those sold for." You compare with remembered examples.
- Approach B: you've worked out a rule of thumb, like "a car loses about 10% of its value each year, plus a bit for every 10,000 km." You apply your model.
Approach A is instance-based learning. Approach B is model-based learning. Both are useful, and in this topic we compare them and then walk through the typical workflow of a real ML project.