FixEval: Execution-based Evaluation of Program Fixes for Competitive Programming Problems

Md Mahim Anjum Haque, Wasi Uddin Ahmad, Ismini Lourentzou, Chris Brown

Keywords: automated program repair, language models, dataset, deep learning, bug fixing

Abstract:

The increasing complexity of software has led to a drastic rise in time and costs for identifying and fixing bugs. Various approaches are explored in the literature to generate fixes for buggy code automatically. However, due to the large combinatorial space of possible fixes for a particular bug, few tools and datasets are available to evaluate model generated fixes effectively. In this work, we introduce FixEval, a benchmark comprising buggy code submissions to competitive programming problems and their respective fixes.
FixEval is composed of a rich test suite to evaluate and assess the correctness of model-generated program fixes and further information regarding time and memory constraints and acceptance based on a verdict. We consider two Transformer language models pretrained on programming languages as our baselines and compare them using match-based and execution-based evaluation metrics. Our experiments show that match-based metrics do not reflect model-generated program fixes accurately. At the same time, execution-based methods evaluate programs through all cases and scenarios designed explicitly for that solution.
Therefore, we believe FixEval provides a step towards real-world automatic bug fixing and model-generated code evaluation. The dataset and models are open-sourced.\footnote{\url{https://github.com/FixEval/FixEval_official}}

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Applications (eg, speech processing, computer vision, NLP)

TL;DR: We introduce FixEval, a novel dataset consisting of competitive programming submissions that incorporates additional program contexts, such as time and space requirements, to evaluate code generated by deep learning models for automatic bug fixing.

Community Implementations: [ 4 code implementations](https://www.catalyzex.com/paper/fixeval-execution-based-evaluation-of-program/code)