2025-11-07 –, Room 301B
Learn how to build fast and reliable retail demand forecasts using StatsForecast, an open-source Python library for scalable statistical forecasting. This session will cover techniques including rolling-origin cross-validation and conformal prediction, with practical retail demand examples.
Retail demand forecasting presents unique challenges: varying seasonal patterns across categories, sharp demand spikes from promotions, and the need to scale across thousands of products efficiently. Traditional statistical models can be accurate but often struggle with speed, scalability, and reproducibility in modern workflows.
This presentation demonstrates how StatsForecast addresses these challenges with faster training, automatic model fitting, and scalable forecasting pipelines built for production. Seamlessly compatible with the Nixtla ecosystem, StatsForecast integrates easily with cross-validation and conformal prediction tools—enabling robust, uncertainty-aware forecasting at scale.
Through hands-on Python examples with retail datasets, you will learn how to:
- Fit Models Automatically: Use AutoARIMA, AutoETS, and other models without manual parameter tuning.
- Scale Efficiently: Train and forecast thousands of series in parallel with Numba-accelerated performance.
- Evaluate Systematically: Apply rolling-origin cross-validation to assess model stability and accuracy.
- Quantify Uncertainty: Combine conformal prediction with statistical models to generate adaptive prediction intervals.
- Integrate Seamlessly: Leverage Nixtla’s unified API for consistent workflows across forecasting tasks.
Khuyen Tran transforms how data scientists learn and work. She is the author of Production-Ready Data Science: From Prototyping to Production with Python, a comprehensive guide that helps data professionals bridge the gap between experimentation and deployment.
As founder of CodeCut, she publishes daily Python tips in her newsletter that reach over 10,000 views per month and has built a community of 110,000 LinkedIn followers.
Previously an MLOps Engineer and Senior Data Engineer at Accenture, she built enterprise data solutions for clients worldwide.
