analyze-results

To analyze ML experiment results across multiple seeds, this skill parses result files from your output directories, computes means and standard deviations for reproducibility, and flags potential outliers or anomalies in your data. It generates a structured comparison table and interpretive insights.

Yes, you can generate a comparison table for different hyperparameters and models. The skill applies analysis across varying models, hyperparameters, and data configurations to reveal performance trends and produce comparable statistics. It helps researchers quickly understand model performance and experiment outcomes.

It works directly with JSON and CSV files located in standard output folders like figures/ and results/, as well as project-specific output directories. The skill automatically identifies and parses these relevant experiment result files to enable structured analysis. You do not need to manually specify file formats.

The best way to flag anomalies in machine learning experiment outputs is to compute statistics across multiple seeds. This skill ensures reproducibility by calculating means and standard deviations, automatically identifying and flagging potential anomalies or outliers within your experiment results. It then generates actionable insights from this analysis.