MLOps.community Real-Time Forecasting Faceoff: Time Series vs. DNNs // Josh Xi // #305
10 snips
Apr 11, 2025 Josh Xi, a Data Scientist at Lyft with a PhD in Operations Research, shares insights on real-time forecasting methods for marketplace dynamics. He discusses the effectiveness of time series models over deep neural networks, emphasizing their adaptability and efficiency. Listeners will learn about integrating external data like weather and events into demand forecasts. Xi also delves into the complexities of geographical forecasting and the role of human insights in enhancing predictive accuracy, making it a must-listen for data enthusiasts!
AI Snips
Chapters
Transcript
Episode notes
Snow's Impact on Taxi Demand
- In 2014, Josh Xi analyzed taxi data and weather, expecting snow to boost demand.
- Surprisingly, heavy snow later had poor correlation, as cities ran out of road salt, hindering travel.
Challenges of External Data
- Acquiring and processing external data, like events and weather, is challenging for real-time forecasting.
- Data pipelines require constant tweaking to improve feature generation and insights.
Time Series vs. DNNs
- Time series forecasting, like autoregression, outperforms DNNs in real-time applications.
- Short-term market dynamics tend to repeat, making historical data highly relevant.