Real-Time Forecasting Faceoff: Time Series vs. DNNs // Josh Xi // #305

Key takeaways

Time series forecasting methods often outperform DNNs in real-time market scenarios due to their simplicity, interpretability, and adaptability.
Integrating external event data can greatly enhance forecasting accuracy, though it comes with challenges, particularly regarding data ingestions and timing.
The cost and infrastructure requirements for training and serving DNNs are significantly higher than those for traditional time series models, impacting operational decisions.
Real-time demand forecasting requires very granular data to allocate drivers effectively, with frequent adjustments based on market conditions.
Future innovations in forecasting may include ensemble models for better accuracy and ongoing adjustments based on historical performance metrics.

Summary

Introduction to Time Series Forecasting vs. DNNs

In this episode, host Demetrios welcomes Josh Xi, a data scientist at Lyft. The conversation kicks off with an exploration of time series forecasting and deep neural networks (DNNs) within the context of real-time market demand and supply forecasting. Josh shares his insights on how Lyft utilizes traditional time series forecasting methods due to their simplicity and effectiveness in comparison to DNNs for specific forecasting tasks.

The Role of External Data in Forecasting

The discussion transitions to the importance of integrating external data sources into forecasting models. Josh explains how events such as sporting events and weather conditions can impact demand and supply forecasting. However, he also highlights the challenges of ingesting and utilizing this data, particularly around uncertainties related to event end times and the variable correlation between weather and demand.

Advantages of Time Series Models

Josh offers a detailed comparison between time series models and DNNs, emphasizing the interpretability and adaptability of time series methods. He explains that time series models can easily incorporate adjustments based on historical data, making them more effective in instances where human intervention is required. The conversation also touches on the cost-efficiency of training traditional models versus DNNs, particularly in real-time forecasting scenarios.

Real-Time Forecasting Challenges and Strategies

The podcast dives deeper into the real-time forecasting challenges faced by Lyft's marketplace team, especially the need for granularity in demand forecasting. Josh discusses how Lyft uses geohashes to segment data and make accurate predictions at a localized level. He also notes the frequent retraining of models to ensure accuracy, underscoring the balance between model complexity and operational costs.

Future Directions and Innovations in Modeling

As the conversation nears its conclusion, Josh reflects on the potential future developments in forecasting models, including the exploration of ensemble methods and advanced parameter tuning for better accuracy. He emphasizes the significance of learning from past forecasting errors and adapting to changing market conditions.

Show notes

Real-Time Forecasting Faceoff: Time Series vs. DNNs // MLOps Podcast #305 with Josh Xi, Data Scientist at Lyft.

Join the Community: https://go.mlops.community/YTJoinIn

Get the newsletter: https://go.mlops.community/YTNewsletter

// Abstract

In real-time forecasting (e.g., geohash level demand and supply forecast for an entire region), time series-based forecasting methods are widely adopted due to their simplicity and ease of training. This discussion explores how Lyft uses time series forecasting to respond to real-time market dynamics, covering practical tips and tricks for implementing these methods, an in-depth look at their adaptability for online re-training, and discussions on their interpretability and user intervention capabilities. By examining these topics, listeners will understand how time series forecasting can outperform DNNs and how to effectively use time series forecasting for dynamic market conditions and decision-making applications.

// Bio

Josh is a data scientist from the Marketplace team at Lyft, working on forecasting and modeling of marketplace signals that power products like pricing and driver incentives. Josh got his PHD in Operations Research in 2013, with minors in Statistics and Economics. Prior to joining Lyft, he worked as a research scientist in the Operations Research Lab at General Motors, focusing on optimization, simulation, and forecasting modeling related to vehicle manufacturing, supply chain, and car-sharing systems.

// Related Links

Website: https://www.lyft.com/

Real-Time Spatial Temporal Forecasting @ Lyft blog: https://eng.lyft.com/real-time-spatial-temporal-forecasting-lyft-fa90b3f3ec24

~~~~~~~~ ✌️Connect With Us ✌️ ~~~~~~~

Catch all episodes, blogs, newsletters, and more: https://go.mlops.community/TYExplore

Join our Slack community [https://go.mlops.community/slack]

Follow us on X/Twitter [@mlopscommunity](https://x.com/mlopscommunity) or [LinkedIn](https://go.mlops.community/linkedin)]

Sign up for the next meetup: [https://go.mlops.community/register]

MLOps Swag/Merch: [https://shop.mlops.community/]

Connect with Demetrios on LinkedIn: /dpbrinkm

Connect with Josh on LinkedIn: /joshxiaominxi

Timestamps: