Forecasting the City in Minutes: Lyft’s Real-Time Spatial-Temporal Forecasting

A deep dive into how Lyft forecasts demand and driver supply across hyper-local geographies (geohash-6) every five minutes, with predictions updated each minute. We explore the trade-offs between fast, updating time-series models and deeper neural networks, why latency and engineering costs matter, and how data characteristics like spatial correlation and geography shape model choice. Learn how these live forecasts power dynamic pricing and driver incentives in a scale-rich marketplace.

Note: This podcast was AI-generated, and sometimes AI can make mistakes. Please double-check any critical information.

Sponsored by Embersilk LLC