Machine Learning with Swift

Implementing KNN in Swift

Fast implementations of KNN and DTW can be found in many machine learning and DSP libraries, for example lbimproved and matchbox C++ libraries:

• github.com/lemire/lbimproved
• github.com/hfink/matchbox

The KNN classifier works with virtually any type of data since you define distance metric for your data points. That's why we define it as a generic structure parameterized with types for features and labels. Labels should conform to a Hashable protocol, as we're going to use them for dictionary keys:

struct kNN<X, Y> where Y: Hashable { ... } 

KNN has two hyperparameters: k—the number of neighbors var k: Int, and distance metric. We'll define it elsewhere, and pass during the initialization. Metric is a function, ...