False Nearest Neighbors Count

Problem

Count false nearest neighbors when increasing the embedding dimension by one. For each point in dimension `dim`, find its nearest neighbor and compare the extra next-coordinate distance to the current distance; count a point when extra / distance > threshold.

Return 0 when parameters are invalid, fewer than two embedding vectors exist, or all distances are zero.

Hint

A high false-neighbor fraction means the embedding dimension is still too small.

Test cases

case 1

in: [[1,1.1,1.2,5,5.1,5.2],1,1,1]

expected: 4

case 2

in: [[1,2,3,4,5,6],1,1,10]

expected: 0

case 3

in: [[1,2,3,4,5],2,1,0.5]

expected: 3

case 4

in: [[5,5,5,5],1,1,1]

expected: 0

Starter codePython

def false_nearest_neighbors_count(series, dim, delay, threshold):
    # Your code here
    pass

4 test cases · run your code against them in the editor