run2-parallel-grid-search

name	run2_parallel-grid-search
description	Parallel DBSCAN hyperparameter grid search with joblib for Mars cloud clustering.

Parallel Grid Search

Optimization Tips

• Pre-group data by image outside the parallel loop
• Use n_jobs=-1 for all cores
• Each evaluation function should be self-contained

from joblib import Parallel, delayed
from itertools import product

params = list(product(min_samples_range, epsilon_range, shape_weight_range))

results = Parallel(n_jobs=-1)(
    delayed(evaluate_params)(ms, eps, sw) for ms, eps, sw in params
)

Greedy Matching for Evaluation

• Compute Euclidean distance matrix between centroids and expert points
• Sort all pairs by distance ascending
• Greedily match closest unmatched pairs (max distance threshold = 100px)
• F1 = 2 * precision * recall / (precision + recall)
  • precision = matched / num_centroids
  • recall = matched / num_expert_points