Reproduce “LExecutor: Learning-Guided Execution” 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
Beatriz Souza, Michael Pradel
LExecutor keeps incomplete Python code running by instrumenting variable, attribute, and function accesses, then using a neural model to predict abstract value classes and inject simple stand-ins when references are missing. On 1,000 open-source functions, CodeT5-guided execution covered a mean 51.6% of non-empty, non-comment lines, versus 22.6% for a frequency-based predictor; coverage does not establish realistic or semantically correct execution.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.