Run1 greedy matching evaluation

Evaluating clustering performance using greedy matching of predicted centroids against ground truth points.From its SKILL.md

Install

npx -y skills add cxcscmu/SkillLearnBench --skill run1_greedy-matching-evaluation

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

SKILL.md

1.1 KB, 227 tokens by cl100k_base, as published. Nobody here has run it

For each image:

Centroid Extraction: Calculate the arithmetic mean of all points assigned to each cluster ID (excluding noise points labeled -1).
Greedy Matching:
- Calculate a distance matrix between all cluster centroids and expert points.
- Iteratively pick the pair with the smallest Euclidean distance (if < 100 pixels).
- Remove these points from the pool and repeat until no pairs are within 100 pixels.
Metrics:
- True Positives (TP): Number of matches found.
- False Positives (FP): Number of unmatched clusters.
- False Negatives (FN): Number of unmatched expert points.
- F1 Score: $2 * TP / (2 * TP + FP + FN)$.
- Delta: Mean standard Euclidean distance of matched pairs.

Ensure you handle images with zero citizen science points or zero expert points correctly by returning F1=0.0 and delta=NaN.

Read from the repository

Just SKILL.md. No reference files, no scripts.