Smooth k-nearest neighbors classifier

Comparison

Results

No cross validation, no fine tuning, Smooth Nearest Neighbors with default hyperparameters, other classfiers with hyperparameters from Classifier comparison, Make Moons, Make Cicles and Linearly Separable as in Classifier comparison, Iris and Wine used as is

	Make Moons	Make Circles	Linearly Separable	Iris	Wine	Average
Smooth Nearest Neighbors	0.95	0.925	0.95	0.9833	0.9722	0.9561
Nearest Neighbors	0.975	0.925	0.95	0.9833	0.9444	0.9556
Gaussian Process	0.975	0.9	0.925	0.9833	0.9583	0.9483
Neural Net	0.9	0.875	0.95	0.9833	0.9861	0.9389
AdaBoost	0.925	0.85	0.95	0.9833	0.9722	0.9361
Decision Tree	0.95	0.775	0.95	0.9833	0.9444	0.9206
Random Forest	0.95	0.75	0.95	0.9833	0.9583	0.9183
Naive Bayes	0.875	0.7	0.95	0.9667	1.0	0.8983
QDA	0.85	0.725	0.925	0.9833	0.9722	0.8911
RBF SVM	0.975	0.875	0.95	0.9833	0.375	0.8317
Linear SVM	0.875	0.4	0.925	0.9	0.9861	0.8172

Installation

pip install smooth-knn

Usage

from smooth_knn.classifier import SmoothKNeighborsClassifier

clf = SmoothKNeighborsClassifier()
clf.fit(X, y)

Blog

https://medium.com/@marek.michalik/adding-smoothness-to-k-nearest-neighbors-4f87b876e8f8