DataFramed #286 Data Science Trends from 2 Kaggle Grandmasters with Jean-Francois Puget, Distinguished Engineer at NVIDIA & Chris Deotte, Senior Data Scientist at NVIDIA
51 snips
Feb 24, 2025 
Join Jean-Francois Puget, a three-time Kaggle Grandmaster and Distinguished Engineer at NVIDIA, along with Chris Deotte, a four-time Kaggle Grandmaster and Senior Data Scientist, as they dive into the transformative role of AI agents in data science. They discuss how GPU acceleration is revolutionizing workflows and the importance of effective model evaluation and communication skills. The conversation also covers the evolving landscape of competitive data science techniques, the future of AI-assisted coding, and the necessity for strong human oversight in this rapidly changing field.
AI Snips
Chapters
Transcript
Episode notes
Few-Shot Learning Impact
- Few-shot learning is gaining traction, reducing the need for large datasets.
- This is particularly effective in text classification, like spam detection, with LLMs.
GPU Acceleration in Data Science
- GPU acceleration is revolutionizing data science by speeding up tools like Pandas, Polars, and Scikit-learn.
- Rapids, a suite of GPU-accelerated libraries, enables faster model training and experimentation.
Brute-Force Feature Engineering
- Chris Deotte used brute-force feature engineering on GPU to win a Kaggle competition.
- He automated testing thousands of feature combinations overnight, significantly boosting his model's accuracy.