Run1 greedy point matching evaluation
Matching predicted cluster centroids to ground truth points using a greedy distance-based approach to calculate F1 score and precision/recall.From its SKILL.md
npx -y skills add cxcscmu/SkillLearnBench --skill run1_greedy-point-matching-evaluationAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
SKILL.md
1.3 KB, 281 tokens by cl100k_base, as published. Nobody here has run it
To evaluate clustering against expert "ground truth" points:
- Centroid Calculation: For each cluster found by DBSCAN, calculate the centroid (mean of $x$ and $y$ coordinates).
- Distance Matrix: Calculate standard Euclidean distance between all predicted centroids and all expert points for a specific image.
- Greedy Matching:
- Find the pair (centroid, expert point) with the smallest Euclidean distance.
- If distance $\le$ threshold (e.g., 100 pixels), count as a True Positive (TP) and remove both points from further matching for this image.
- Repeat until no more pairs can be matched under the threshold.
- Metrics:
- TP: Number of matched pairs.
- FP (False Positives): Number of unmatched predicted centroids.
- FN (False Negatives): Number of unmatched expert points.
- F1 Score: $2 \cdot TP / (2 \cdot TP + FP + FN)$. If $TP+FP+FN = 0$, F1 is typically 1.0 (though in this task, if no points exist/match, the requirement specifies 0.0).
- Delta: Average Euclidean distance of all matched TP pairs.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.