knnFeat

Feature Extraction with KNN

Description

Python implementation of feature extraction with KNN.

And @momijiame updated my implementation. I recommend to use this:
https://github.com/momijiame/gokinjo

pip install gokinjo

The following is R implementation:
http://davpinto.com/fastknn/articles/knn-extraction.html#understanding-the-knn-features

Requirements

• Python 3.x
  • numpy
  • scikit-learn
  • scipy

Install

git clone git@github.com:upura/knnFeat.git
cd knnFeat
pip install -r requirements.txt

Demo

Notebook version can be seen here.

Packages for visualization

import numpy as np
%matplotlib inline
import matplotlib.pyplot as plt

Data generation

x0 = np.random.rand(500) - 0.5
x1 = np.random.rand(500) - 0.5
X = np.array(list(zip(x0, x1)))
y = np.array([1 if i0 * i1 > 0 else 0 for (i0, i1)  in list(zip(x0, x1))])

Visualization

Feature extraction with KNN

from knnFeat import knnExtract
newX = knnExtract(X, y, k = 1, folds = 5)

Visualization

Algorithm

Quote from here.

It generates k * c new features, where c is the number of class labels. The new features are computed from the distances between the observations and their k nearest neighbors inside each class, as follows:

1. The first test feature contains the distances between each test instance and its nearest neighbor inside the first class.
2. The second test feature contains the sums of distances between each test instance and its 2 nearest neighbors inside the first class.
3. The third test feature contains the sums of distances between each test instance and its 3 nearest neighbors inside the first class.
4. And so on.

This procedure repeats for each class label, generating k * c new features. Then, the new training features are generated using a n-fold CV approach, in order to avoid overfitting.

Development

flake8 .

pytest

pytest -v -m 'success' --cov=.

License

MIT

Author

upura