Reproduce “Coding LLMs Can Be Harnessed to Learn Interpretable Models from Data” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Aditya Kumar, Cassandra A. Cohen, William W. Cohen

CLIPPER adapts SLIPPER's rule-learning loop: a coding model proposes executable rules from weighted samples, which the harness scores on training data and combines. Saved rules classify without an LLM call at inference. On seven tabular tasks at 1,200 examples, CLIPPER achieves 22.5% macro test error versus SLIPPER's 23.4%, with both limited to 100 examples per proposal.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.