evo-mars-clustering-eval

Core clustering and evaluation utilities for DBSCAN-based Mars cloud annotation clustering.

Key Functions

• weighted_euclidean_pairwise(u, v, w) - Custom distance: sqrt((w*dx)^2 + ((2-w)*dy)^2)
• compute_precomputed_distance_matrix(X, w) - Build precomputed distance matrix for DBSCAN
• run_dbscan_and_get_centroids(X, eps, min_samples, dist_matrix) - Run DBSCAN, return centroids
• greedy_match(centroids, ground_truth, threshold=100.0) - Greedy bipartite matching
• score_image(citsci_points, expert_points, eps, min_samples, w) - Full per-image scoring

Usage

import sys
sys.path.insert(0, '/app/environment/skills/evo-mars-clustering-eval/scripts')
from utils import score_image, greedy_match, run_dbscan_and_get_centroids

f1, delta = score_image(citsci_xy, expert_xy, eps=10, min_samples=5, w=1.0)

Domain Rules

• DBSCAN uses precomputed distance matrix with custom weighted Euclidean metric
• Greedy matching uses standard Euclidean distance (not custom), max threshold 100px
• F1 = 2TP / (2TP + FP + FN), handles edge cases (both empty = 1.0)
• Delta = mean standard Euclidean distance of matched pairs, NaN if no matches
• Noise points (label -1) are excluded from centroid computation