What we talk about when we talk about scientific programming
summary
Updates to the same remote dataset change land‑cover conclusions dramatically — an example shows deforestation jumping from 3% to 8.5% when the provider updates their analysis — and that fragility breaks reproducibility. Workflows rely on Google Earth Engine, container tools like Docker and Nyx, and ad hoc shell scripts, so a prototype called shelter records sessions, uses eBPF dependency tracking, supports safe merges and undo, and is work‑in‑progress rather than a finished reproducibility solution.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 1:28Scientific programming defined
- 3:15Data updates and reproducibility
- 6:49Code and research software
- 9:39Empathy for vernacular programmers
- 10:45Shelter tool for reproducible shells
- 12:33Q&A: Exploratory programming
- 14:54Q&A: Future goals
related talks
01h 11mProgramming for the planet
20mModernizing the Data Science Toolkit of a 40-year-old Market Research Company
41mSoftware environmentalism
01h 00mLarge-Scale Changes at Google: Lessons Learned From 5 Yrs of Mass Migrations
39mApplying refactoring techniques ...
38mBridging the Chasm Between Research & Software Development
31mKeynote – Amanda Casari
24mIt's Time To Rethink Everything