PyData London 2025 - Day2
PyData London 2025 - Day2
Personal highlights:
🎤 Opening speech by our Conference Chair John Carney and director of NumFOCUS Dr. R. Terrel (he/him) .
📌Keynote - Keep Calm and Data On: Being a data science practitioner in the era of AI proliferation by Leanne Kim Fitzpatrick
Leanne discussed the impact of the AI arms race on the field, the reality of being a practitioner, and principles for navigating the current situation.
📌Keynote - From Next Token Prediction to Reasoning and Beyond by Jay Alammar.
Jay provided an overview of LLM concepts for data professionals, including Transformer architectures, tokenisers, reward models, reasoning LLMs, agentic trajectories, and training stages.
🧵 Parallel PyTorch Inference with Python Free-Threading by Michał Szołucha Key takeaways:
- Python has (almost) removed GIL from 3.13.
- GPU can run faster.
- So can/will PyTorch.
🌐 Sovereign Data for AI with Python by Lex Avstreikh I have learnt:
- What is sovereign data.
- Challenges, opportunities and strategies for managing sovereign data.
- The platform/solution Hopsworks provides.
⚽ Cutting Edge Football Analytics using Polars, Keras and Spektral by Joris Bekkers Good to see:
- A GNN model for football analytics.
- The cool open source library of kloppy for standardised football data.
🧪PyScript - Python in the Browser by Chris Laffra. Highlights: -PyScript is a fast-growing and vibrant open-source platform for Python in the browser.
- Multiple examples of using PyScript.
- I need to check it!
🕸️ NetworkX is Fast Now: Zero Code Change Acceleration by Mridul Seth Glad to hear:
- NetworkX is popular and people love it (I certainly do!)
- NetworkX too slow for hugh graphs? Accelerate its backend, there are many options. Such as GraphBLAS and nx-cugraph
And of course, the happy hour after all talks! 🍻