Reproduce “Compositional API Recommendation for Library-Oriented Code Generation” 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
Zexiong Ma, Shengnan An, Bing Xie, Zeqi Lin
CAPIR decomposes multi-step library requests into subtasks, retrieves APIs for each, and reranks combined candidates. On Torchdata-AR, at k=5, it raises Recall@5 from 18.7% to 43.2% and Precision@5 from 15.5% to 37.1% versus adapted CLEAR. On Torchdata-Code, GPT-3.5 pass@100 rises from 16% to 28% with CAPIR; code generation was evaluated only in Python, and Torchdata was the only low-resource library.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.