data-matching

name	data-matching
description	Greedy matching of centroid points to expert annotations based on standard Euclidean distance.

Logic

Greedy matching for two point sets.

import numpy as np
from scipy.spatial.distance import cdist

def match_points(centroids, expert_points, max_dist=100.0):
    if len(centroids) == 0 or len(expert_points) == 0:
        return 0, 0, 0 # TP, FN, FP
    
    dist_matrix = cdist(centroids, expert_points, 'euclidean')
    # Find all pairs within distance
    matches = []
    for i in range(len(centroids)):
        for j in range(len(expert_points)):
            if dist_matrix[i, j] <= max_dist:
                matches.append((dist_matrix[i, j], i, j))
    
    matches.sort(key=lambda x: x[0]) # Closest first
    
    used_centroids = set()
    used_experts = set()
    total_dist = 0
    num_matches = 0
    
    for dist, i, j in matches:
        if i not in used_centroids and j not in used_experts:
            used_centroids.add(i)
            used_experts.add(j)
            total_dist += dist
            num_matches += 1
            
    # TP: num_matches
    # FN: len(expert_points) - num_matches
    # FP: len(centroids) - num_matches
    return num_matches, len(expert_points) - num_matches, len(centroids) - num_matches, total_dist