What problem does it solve?
Manually processing scattered ML experiment results, calculating performance statistics, and comparing model performance across multiple runs or hyperparameter configurations is time-consuming and prone to human error, especially when dealing with large numbers of trials or multi-seed reproducibility checks.
Core Features & Use Cases
- Automated Result Parsing: Automatically locate and parse JSON/CSV result files from common project output directories like
figures/ or results/.
- Statistical Analysis: Calculate mean and standard deviation for multi-seed runs, identify trends in parameter sweeps, and flag anomalous or suspicious results.
- Structured Insight Generation: Produce actionable findings with clear observations, interpretations, and suggested next experiments to advance research.
- Use Case: If you ran 10 training runs of different vertebrae segmentation models for your MICCAI 2025 submission, this skill will automatically compile Dice scores, calculate improvements over your baseline model, and highlight the best-performing configuration.
Quick Start
Use the analyze-results skill to compare the performance of all model checkpoints in the ./experiment_outputs folder and generate a summary of key findings and recommended next steps.
