Multiresolution Learning Dynamics Experiments

This repository contains the cleaned derived data, analysis scripts, generated figures, and raw-pipeline provenance code for the paper.

Repository: https://github.com/jurij-jukic/key_search_paper_code

Heavyweight model checkpoints and panel reconstruction dumps are not included. The cleaned metrics here are sufficient to reproduce the paper figures and reported numerical summaries.

Contents

• data/: cleaned CSV/JSON artifacts used by the analysis scripts.
• results/: derived tables used for the reported statistics.
• figures/: generated figure outputs used by the paper.
• *.py: scripts that regenerate the reported figures and tables.
• raw_pipeline/: provenance-only raw training and artifact processing code, included for inspection but not required for lightweight reproduction.

Reproducing Derived Artifacts

Use a fresh virtual environment rather than a system Python with pre-existing user-site packages.

python3 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install -r requirements.txt

python make_loss_figures.py
python make_weibull_figures.py
python compute_weibull_tables.py
python compute_clock_law.py

The scripts write outputs under results/ and figures/.

Environment

The analysis scripts require Python 3.9+ with NumPy, Pandas, SciPy, and Matplotlib.