DataFramed

#356 The Forecast for Time Series Forecasts with Rami Krispin, Senior Manager of Data Science at Apple

54 min

This episode explores the evolution of time series forecasting, focusing on the emergence of foundation models and their ability to handle large-scale datasets that traditional methods struggle with. Rami Krispin from Apple discusses the critical aspects of getting forecasts into production, emphasizing risk management, the importance of feature engineering for real-world events, and transparent communication of model limitations and uncertainties to build trust.

Summarized by Podsumo

🎧 Listen 🎙️ Ask about this episode

✨ Key Takeaways

1

Traditional time series models struggle with the exponential growth of data, making foundation models crucial for handling thousands of time series (e.g., Walmart's product demand) at scale.
2

Getting time series models into production requires robust data pipelines, continuous monitoring for data drift, and a clear strategy for risk management, especially for critical forecasts like energy capacity.
3

While foundation models are good at identifying outliers, predicting future rare or non-seasonal events (like major concerts or policy changes) remains a significant challenge, often requiring extensive feature engineering or augmentation with external information.
4

Building trust in forecasts involves transparent communication of model limitations, using prediction intervals to convey uncertainty, and explaining model decisions to non-technical stakeholders.
5

The future of time series forecasting sees data scientists evolving into 'architects' or 'designers,' leveraging automated tools and foundation models, but still requiring a strong understanding of underlying statistics and engineering principles.

💬 Notable Quotes

"I'm not sure if people would feel comfortable yet to give the power of those decision making for autonomous agents."
"When it's doing great, it's great. But you need to understand that when you're using auto models, auto algorithms, that sometimes you get it wrong."
"Statistician are very set people. They know that they're wrong from the get-go and they go and measure it."