This project is a super simple implementation of a cpp library. In this case, I decided to implement a simple algorithm such as the K-Nearest Neighbor Algorithm. With this project, rather than work on a complex project, I decided to build a simple "Machine Learning" library to perform KNN inference in order to validate some OOP concepts that I have been learnign recently as well as my knowledge on how to structure a cpp project.

The k-nearest neighbor algorithm is a non-parametric (aka being distribution-free) classification algorithm that is used for both regression and classification.

The idea behind the algorithm is really simple and it shouldn't be confused with the k-means clustering algorithm, even though some clustering might be considered in this case. The KNN algorithm predicts the label of class of a data point based on the label of its K nearest neighbors, so in some sense it builds a cluster with the K nearest neighbours. The predominant class in that cluster will be given to the unknown data point, so, an object is classified by a plurarity vote of its neighbors and it's given the most common class in that neighborhood.

It can use different types of norms to measure the distance between the neighbors, but in this case I have decided to rely on the basics and use the euclidean norm aka l2 norm.

So, as I have stated several times, it is a simple version of the algorithm and the Knn object (which is thought to be a 'model' object) contains the following attributes and methods:

• Private methods:

  • euclidean_distance method, to get the l2 norm with all its neighbors.
  • get_knn method, which gets the information about the k nearest neighbors based on the euclidean distance between them and the POI.

• Public methods:

  • predict_class method, which is the main method of the ones implemented and is the one which returns the inferred class.

• Public attributes:

  • neighbours_number, which is basically k and defines the number of nearest neighbors to use.

The src/main.cpp shows an example on how to use the 'library' (I don't think is complex/big/useful enough to be called so). It uses some data for the well-known Iris Flower Dataset.

There are a lot of things that I may add to this simple project, but one of them would be to build a more general model for the training dataset.