AI Efficiency in Science - Replication Package

This repository contains the data and code necessary to replicate the findings, figures, and supplemental analyses of the paper on AI efficacy and efficiency in scientific applications.

Repository Structure

• paper/
  • data/: Contains processed_db.csv, the primary dataset of 2,507 comparisons.
  • src/: Python scripts for generating the main paper figures (Figures 2–6).
  • figs/: Output directory for generated figures (PNG and SVG formats).
  • figs/data/: Data exports used to populate the tables and figures.
  • supplementary/:
    • data/: Contains supplemental_report.md (consolidated statistical results) and metric mappings.
    • src/: generate_supplemental_report.py handles all supplementary statistical calculations.

Getting Started

Prerequisites

Ensure you have Python 3.8+ and the following libraries installed:

pip install pandas matplotlib seaborn numpy statsmodels scipy

Running the Replication

To regenerate all figures and the supplemental report, run the provided shell script from the repository root:

./run_all.sh

This script will:

1. Generate every main-text figure.
2. Produce the consolidated supplemental report (paper/supplementary/data/supplemental_report.md).
3. Verify that all outputs match the database.

Methodology Note

The core logic for performance-cost normalization and quadrant classification is centralized in paper/supplementary/src/generate_supplemental_report.py. This ensures consistency between the directional bar charts, radial scatter plots, and robustness checks.

---

Note: Any directories named old/ contain legacy scripts or temporary files and can be ignored for replication purposes.