analyze-results

To analyze machine learning experiment results, the Skill parses JSON and CSV output files to compute statistics, generate comparison tables, and extract actionable insights from your metrics. It automatically locates results in common project directories and structures them for analysis.

Comparing model metrics against a baseline across hyperparameter sweeps involves grouping results by independent variables and computing deltas. The Skill organizes these variables, identifies sweep trends, and reports differences to highlight which configurations outperform the baseline.

Reporting mean and standard deviation for multi-seed reproducibility requires aggregating runs and flagging outliers. The Skill calculates mean±std for multi-seed runs, identifies anomalies, and translates statistical findings into clear observation and implication narratives.

Yes, generating actionable next-step recommendations from raw experiment logs is a core feature. The Skill converts raw metrics into a structured data table with numbered key findings and suggests specific follow-up experiments to guide your research.

No specific dependencies or frameworks are required to perform statistical analysis on experiment outputs. The Skill operates independently to parse your JSON and CSV files and compute statistical deltas without needing external libraries or environment setups.

If experiment results are scattered across different directories, the Skill handles result discovery by searching common project folders. It locates relevant JSON and CSV outputs across your workspace and aggregates them into a structured format for comparison.