Bug Analysis in Jupyter Notebook Projects: An Empirical Study

Computational Notebooks, such as Jupyter, have been widely adopted in data science for building data-driven code. Despite their popularity, challenges related to software development in these environments still need to be investigated. This study conducts a systematic analysis of bugs and difficulties faced by Jupyter practitioners. A total of 14,740 commits from 105 GitHub projects were mined, and 30,416 Stack Overflow posts were analyzed to identify common issues. Additionally, 19 interviews with data scientists were conducted to gather more details on these challenges. For validation, a survey with various professionals was carried out, along with an analysis based on the Apriori algorithm. Based on these findings, a taxonomy of bugs was proposed to classify different types of issues found in Jupyter projects.