Machine Learning With Go

kNN assumptions and pitfalls

Due to its simplicity, kNN does not have too many assumptions. However, there are some common pitfalls that you should be aware of as you apply kNN:

• kNN is evaluated lazily. By this, we mean that the distances or similarities are calculated when we need to make a prediction. There is not really anything to train or fit prior to making a prediction. This has some advantages, but the calculation and search over points can be slow when you have many data points.
• The choice of k is up to you, but you should put some formalism around choosing k and provide justification for the k that you choose. A common technique to choose k is just to search over a range of k values. You could, for example, start with k = 2. Then, ...