Project Zouwu: Scalable AutoML for Telco Time Series Analysis using Ray and Analytics Zoo
summary
Project Zouwu delivers scalable AutoML for telco time series by wiring Ray Tune on Ray into Analytics Zoo so feature generation, model selection and hyperparameter search run distributed across Spark, Kubernetes or HDFS clusters. A time-sequence predictor drives trials that log results to HDFS and produce a savable pipeline. Applied to SK Telecom KPI forecasting (use past week to predict next two hours) it outperforms AR/ETS baseline and feeds anomaly alerts in production.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 0:49Background and Analytics Zoo
- 3:35Time Series Forecasting Basics
- 5:44AutoML and Ray Tune Integration
- 7:11Time Series AutoML Framework
- 12:37Project Zouwu Overview
- 13:40User Case Studies
- 16:52Future Work and Conclusion
related talks
22mScalable AutoML for Time Series Forecasting
22mHow Zoox Accelerated Autonomous Driving with Ray
30mReal-Time ML services: From ideation to production
50mAutomated data analyzer using Machine Learning
28mScaling AI & Machine Learning Workloads With Ray on AWS, Kubernetes, & BERT
29mScaling ML Pipelines for Autonomous Driving with Ray
20mThe Quick Journey to Using Ray: How We Implement Ray and Anyscale to Speed up our ML Processes
44mPanel: ML Platform on Ray