Suggesting Code Edits in Interactive Machine Learning Notebooks Using Large Language Models
Bihui Jin, Jiayue Wang, Pengyu Nie
TL;DR
This work introduces the first large-scale dataset of Jupyter notebook edits in ML repositories (48,398 edits across 792 projects) to study real-world maintenance and support large language models in predicting code edits. It provides a detailed data collection pipeline, extensive statistics, and splits for training, validation, and testing. Through few-shot prompting and finetuning of DeepSeek-Coder models on file- and cell-level edits, the study reveals that, despite larger models and context-rich prompts, predicting realistic notebook edits remains challenging, underscoring the need for advanced techniques like retrieval-augmented generation and agentic systems. The dataset and code are open-sourced to enable replication and further research on ML code maintenance using LLMs.
Abstract
Machine learning developers frequently use interactive computational notebooks, such as Jupyter notebooks, to host code for data processing and model training. Jupyter notebooks provide a convenient tool for writing machine learning pipelines and interactively observing outputs, however, maintaining Jupyter notebooks, e.g., to add new features or fix bugs, can be challenging due to the length and complexity of the notebooks. Moreover, there is no existing benchmark related to developer edits on Jupyter notebooks. To address this, we present the first dataset of 48,398 Jupyter notebook edits derived from 20,095 revisions of 792 machine learning repositories on GitHub, and perform the first study of the using LLMs to predict code edits in Jupyter notebooks. Our dataset captures granular details of cell-level and line-level modifications, offering a foundation for understanding real-world maintenance patterns in machine learning workflows. We observed that the edits on Jupyter notebooks are highly localized, with changes averaging only 166 lines of code in repositories. While larger models outperform smaller counterparts in code editing, all models have low accuracy on our dataset even after finetuning, demonstrating the complexity of real-world machine learning maintenance tasks. Our findings emphasize the critical role of contextual information in improving model performance and point toward promising avenues for advancing large language models' capabilities in engineering machine learning code.