wandb-analysis

What problem does it solve?

Manually reviewing and comparing Weights & Biases experiment runs to extract metrics, verify results, and generate analysis outputs is slow, error-prone, and difficult to scale when working with large numbers of runs, sweeps, or ablation studies.

Core Features & Use Cases

• Run Comparison & Metric Extraction: Programmatically query the W&B API to pull metrics across runs, compare hyperparameters, seeds, or method variants for paper tables, leaderboards, or internal reporting.
• Ablation & Sweep Analysis: Isolate component contributions via grouped ablation runs, run statistical significance tests across seeds, and identify top-performing hyperparameter configurations from sweeps.
• Result Verification & Export: Cross-check reported metrics against actual logged W&B values, export run data to CSV/JSON for offline plotting, and validate git provenance to ensure results match the correct code version.

Quick Start

Use the wandb-analysis skill to pull final accuracy metrics for all finished runs in your W&B project and generate a sorted comparison table for your experiment report.