Generative Code Modeling with Graphs

TL;DR

This paper introduces a novel graph-based generative model for source code that integrates grammar-driven expansion with neural message passing, effectively capturing syntax and semantics to produce meaningful programs.

Contribution

The paper presents a new graph-based generative approach that combines grammar rules with neural message passing for source code generation, improving semantic correctness.

Findings

Outperforms strong baseline models in code generation tasks

Generates semantically meaningful expressions

Effectively captures syntax and semantics in code generation

Abstract

Generative models for source code are an interesting structured prediction problem, requiring to reason about both hard syntactic and semantic constraints as well as about natural, likely programs. We present a novel model for this problem that uses a graph to represent the intermediate state of the generated output. The generative procedure interleaves grammar-driven expansion steps with graph augmentation and neural message passing steps. An experimental evaluation shows that our new model can generate semantically meaningful expressions, outperforming a range of strong baselines.