What problem does it solve?
Interpreting raw machine learning experiment outputs scattered across JSON and CSV files is tedious and error-prone, making it hard to compare runs, compute deltas against baselines, and draw defensible conclusions.
Core Features & Use Cases
- Result Aggregation: Locates and parses JSON/CSV result files from figures/, results/, or project output directories into structured data.
- Comparison Tables: Organizes results by independent variables (model type, hyperparameters, data config) and dependent metrics, always computing delta versus baseline.
- Statistical Analysis: Reports mean and standard deviation across seeds, identifies parameter sweep trends, and flags outliers or suspicious results.
- Use Case: After running a learning-rate sweep with three seeds, ask the assistant to analyze the results directory and receive a comparison table, key findings with interpretations, and suggested follow-up experiments.
Quick Start
Analyze the experiment results in my results/ directory and compare them against the baseline model.