Lightning talks - session 2
summary
A new library called FireDuck claims a pandas-compatible compiler with lazy JIT that is multi‑threaded rather than a thin wrapper, and it performs query planning and IR optimizations before execution. The project reports a real case where a cloud job drops from 3.5 hours to 10 minutes, cutting cost and emissions. It accepts normal pandas code, integrates with matplotlib/plotly, and produces compiled, parallel execution plans rather than executing Python row by row.
AI-written summary · about
chapters (16)
- 0:00Lightning talks intro
- 1:28Fire Duck high-performance pandas
- 5:19AI robustness and adversarial attacks
- 8:33PyLadies Paris community
- 11:47Python for mapping holiday plans
- 14:57PyCafe interactive Python UIs
- 17:04ADHD in tech and neurodiversity
- 20:15Language ideology and AI uniformity
- 24:08AnyWidget standard for UI components
- 26:40Composable functions workflows
- 29:48Optimism through data
- 33:22Robust AutoML with tuning
- 35:57Reviving MacBook Pro with DIY
- 39:53JupyterLab AI chat extension
- 43:25AI agents reading French laws
- 48:30Python versions in Conda Forge
related talks
01h 02mLightning Talks (02) & Closing Notes
35mHow many dataframe libraries do you need to change a lightbulb?
32mLightning Talks
38mLightning Talks
45mPyData London 2018 Lightning Talks - Day 2
26mLife Beyond Pandas: Workflows with DuckDB, Daft, Polars, and Datafusion
48mLightning Talks
53mPyData London 2018 Lightning Talks - Day 1