simple outlier detection using kNN

knn_test.py

#!/usr/bin/env python3

from sklearn.neighbors import NearestNeighbors
import numpy as np
import sys

th_pi = 20
th_ep = 5.0

def main():
    dataset = []
    with open(sys.argv[1]) as f:
        for line in f:
            dataset.append([float(line)])
    num_entries = len(dataset)
    X = np.array(dataset)
    nbrs = NearestNeighbors(n_neighbors=num_entries-1, algorithm='ball_tree').fit(X)
    for i in dataset:
        distances, indices = nbrs.kneighbors(np.array([i]))
        count = 0
        for d in distances[0]:
            if d <= th_ep:
                count += 1
        if 100 * count // num_entries <= th_pi:
            print("outlier ->", i)

main()