Exploration and Explanation in Computational Notebooks
summary
Analysis of over 1 million public Jupyter notebooks on GitHub finds 27% have no explanatory text and a median of ~20 cells and ~100 lines of code. A curated set of 145 notebooks for 50 academic papers shows 77% include text but 88% only describe steps while about one third discuss reasoning or results. Interviews with 15 academic analysts reveal notebooks are treated as personal and messy, and simple defaults like flipping new cells to markdown are proposed to nudge explanation.
AI-written summary · about
chapters (11)
- 0:00Introduction and motivation
- 1:09Lessons from Reinhart-Rogoff paper
- 2:30Computational notebooks overview
- 3:52Research questions and methods
- 4:34Analysis of GitHub notebooks
- 7:21Academic notebooks study
- 9:09Text usage in academic notebooks
- 10:11Interviews with analysts
- 12:57Exploration vs explanation tension
- 13:39Summary and future directions
- 15:04Q&A session
related talks
17mNotes on Notebooks: Is Jupyter the Bringer of Jollity?
33mSignificant Roadblocks to Usefulness for Jupyter Notebooks and a Recipe to Fix them
25mThe First Notebook War
48mWhither Problem-Solving Environments?
28mUnlock the Full Potential of Jupyter Notebooks
34mCode the Docs: Interactive Document Environments
34mReadme.md? We Can Do Better
49mBest tools for Jupyter notebooks