An official website of the United States government
Software developers create and share code online to demonstrate programming language concepts and programming tasks. Code snippets can be a useful way to explain and demonstrate a programming concept, but may not always be directly executable. A code snippet can contain parse errors, or fail to execute if the environment contains unmet dependencies. This paper presents an empirical analysis of the executable status of Python code snippets shared through the GitHub gist system, and the ability of developers familiar with software configuration to correctly configure and run them. We find that 75.6% of gists require non-trivial configuration to overcome missing dependencies, configuration files, reliance on a specific operating system, or some other environment configuration. Our study also suggests the natural assumption developers make about resource names when resolving configuration errors is correct less than half the time. We also present Gistable, a database and extensible framework built on GitHub's gist system, which provides executable code snippets to enable reproducible studies in software engineering. Gistable contains 10,259 code snippets, approximately 5,000 with a Dockerfile to configure and execute them without import error. Gistable is publicly available at https://github.com/gistable/gistable.
more »
« less
CoaCor: Code Annotation for Code Retrieval with Reinforcement Learning
To accelerate software development, much research has been performed to help people understand and reuse the huge amount of available code resources. Two important tasks have been widely studied: code retrieval, which aims to retrieve code snippets relevant to a given natural language query from a code base, and code annotation, where the goal is to annotate a code snippet with a natural language description. Despite their advancement in recent years, the two tasks are mostly explored separately. In this work, we investigate a novel perspective of Code annotation for Code retrieval (hence called “CoaCor”), where a code annotation model is trained to generate a natural language annotation that can represent the semantic meaning of a given code snippet and can be leveraged by a code retrieval model to better distinguish relevant code snippets from others. To this end, we propose an effective framework based on reinforcement learning, which explicitly encourages the code annotation model to generate annotations that can be used for the retrieval task. Through extensive experiments, we show that code annotations generated by our framework are much more detailed and more useful for code retrieval, and they can further improve the performance of existing code retrieval models significantly.
more »
« less
- Award ID(s):
- 1815674
- PAR ID:
- 10106912
- Date Published:
- Journal Name:
- The World Wide Web Conference
- Page Range / eLocation ID:
- 2203 to 2214
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Abstract Source code is a form of human communication, albeit one where the information shared between the programmers reading and writing the code is constrained by the requirement that the code executes correctly. Programming languages are more syntactically constrained than natural languages, but they are also very expressive, allowing a great many different ways to express even very simple computations. Still, code written by developers is highly predictable, and many programming tools have taken advantage of this phenomenon, relying on language modelsurprisalas a guiding mechanism. While surprisal has been validated as a measure of cognitive load in natural language, its relation to human cognitive processes in code is still poorly understood. In this paper, we explore the relationship between surprisal and programmer preference at a small granularity—do programmers prefer more predictable expressions in code? Usingmeaning‐preserving transformations, we produce equivalent alternatives to developer‐written code expressions and run a corpus study on Java and Python projects. In general, language models rate the code expressions developerschooseto write as more predictable than these transformed alternatives. Then, we perform two human subject studies asking participants to choose between two equivalent snippets of Java code with different surprisal scores (one original and transformed). We find that programmersdoprefer more predictable variants, and that stronger language models like the transformer align more often and more consistently with these preferences.more » « less
-
Recently, there have been significant advances and wide-scale use of generative AI in natural language generation. Models such as OpenAI’s GPT3 and Meta’s LLaMA are widely used in chatbots, to summarize documents, and to generate creative content. These advances raise concerns about abuses of these models, especially in social media settings, such as large-scale generation of disinformation, manipulation campaigns that use AI-generated content, and personalized scams. We used stylometry (the analysis of style in natural language text) to analyze the style of AI-generated text. Specifically, we applied an existing authorship verification (AV) model that can predict if two documents are written by the same author on texts generated by GPT2, GPT3, ChatGPT and LLaMA. Our AV model was trained only on human-written text and was effectively used in social media settings to analyze cases of abuse. We generated texts by providing the language models with fanfiction snippets and prompting them to complete the rest of it in the same writing style as the original snippet. We then applied the AV model across the texts generated by the language models and the human written texts to analyze the similarity of the writing styles between these texts. We found that texts generated with GPT2 had the highest similarity to the human texts. Texts generated by GPT3 and ChatGPT were very different from the human snippet, and were similar to each other. LLaMA-generated texts had some similarity to the original snippet but also has similarities with other LLaMA-generated texts and texts from other models. We then conducted a feature analysis to identify the features that drive these similarity scores. This analysis helped us answer questions like which features distinguish the language style of language models and humans, which features are different across different models, and how these linguistic features change over different language model versions. The dataset and the source code used in this analysis have been made public to allow for further analysis of new language models.more » « less
-
Instruction tuning is a supervised fine-tuning approach that significantly improves the ability of large language models (LLMs) to follow human instructions. For programming tasks, most models are finetuned with costly human-annotated instruction-response pairs or those generated by large, proprietary LLMs, which may not be permitted. We propose SelfCodeAlign, the first fully transparent and permissive pipeline for self-aligning code LLMs without extensive human annotations or distillation. SelfCodeAlign employs the same base model for inference throughout the data generation process. It first extracts diverse coding concepts from high-quality seed snippets to generate new tasks. It then samples multiple responses per task, pairs each with test cases, and validates them in a sandbox environment. Finally, passing examples are selected for instruction tuning. In our primary experiments, we use SelfCodeAlign with CodeQwen1.5-7B to generate a dataset of 74k instruction-response pairs. Finetuning on this dataset leads to a model that achieves a 67.1 pass@1 on HumanEval+, surpassing CodeLlama-70B-Instruct despite being ten times smaller. Across all benchmarks, this finetuned model consistently outperforms the original version trained with OctoPack, the previous state-of-the-art method for instruction tuning without human annotations or distillation. Additionally, we show that SelfCodeAlign is effective across LLMs of various sizes, from 3B to 33B, and that the base models can benefit more from alignment with their own data distribution. We further validate each component’s effectiveness in our pipeline, showing that SelfCodeAlign outperforms both direct distillation from GPT-4o and leading GPT-3.5-based distillation methods, such as OSS-Instruct and Evol-Instruct. SelfCodeAlign has also led to the creation of StarCoder2-Instruct, the first fully transparent, permissively licensed, and self-aligned code LLM that achieves state-of-the-art coding performance. Overall, SelfCodeAlign shows for the first time that a strong instruction-tuned code LLM can result from self-alignment rather than distillation.more » « less
-
Code search is vital in the maintenance and extension of software systems. Past works have used separate language models for the natural language and programming language artifacts on models with multiple encoders and different loss functions. Similarly, this work approaches code search for Python as a translation retrieval problem while the natural language queries and the programming language are treated as two types of languages. By using dual encoders, these two types of language sequences are projected onto a shared embedding space, in which the distance reflects the similarity between a given pair of query and code. However, in contrast to previous work, this approach uses a unified language model, and a dual encoder structure with a cosine similarity loss function. A unified language model helps the model take advantage of the considerable overlap of words between the artifacts, making the learning much easier. On the other hand, the dual encoders trained with cosine similarity loss helps the model learn the underlining patterns of which terms are important for predicting linked pairs of artifacts. Evaluation shows the proposed model achieves performance better than state-of-the-art code search models. In addition, this model is much less expensive in terms of time and complexity, offering a cheaper, faster, and better alternative.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1145/3308558.3313632
