K-Nearest-Neighbours (KNN) from Scratch with Julia

This repo implements the KNN classification and regression model in pure Julia code. The objective is to demonstrate the practical implementation of the model to aid the learning process of anyone.

We suggest using the NearestNeighbors.jl package for serious use of this model in real applications with Julia.

Inspired by Scikit-Learn implementation of KNN, you just need to pass the data as Matrix in X argument, the variable to predict like Vector in y argument and the number of K to fit the model. Also, to make a prediction use the predict() function adding the new data in arguments.

# package to import data
using RDatasets, DataFrames

# prepare data to model
df = dataset("datasets", "iris");
X = Array(df[:, 1:end-1]);
y = Array(df[:, end]);

# load the knn model
include("knn_functions.jl")

# fit the model
K = 5
model = KnnClassifier(X, y, K);

# predict new data
predict(model, rand(4))         # 1 observation
predict(model, rand(10, 4))     # multiple observations

This implementation is faster than the Scikit-Learn model, showed in the pictures below, mainly because of two reasons:

• It's has been created with Julia with all the good practices to extract the maximum performance possible of the lang.
• That it's an light implementation, just with the minimum requirements to use a KNN model, with train and prediction.

Scikit-Learn with C in backend

This Julia implementation