Integrating a Spark ML Pipeline into a High-Load Low-Latency Recommendation Service
summary
The talk explains how to deploy a Spark ML Pipeline trained on large-cluster data into a high-throughput, low-latency online recommendation service. It focuses on using Apache Spark (with some SynapseML) and code in Scala and Java to meet strict response-time requirements for production inference.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 2:15Problem and tools
- 6:25Recommendation service
- 8:15Spark ML integration issue
- 9:21Microsoft library solution
- 12:12Streaming problems and limits
- 15:23Optimization via internal interface
- 18:38Custom data source
- 22:44Connector and API
- 24:55Configuration and timeouts
- 28:52Results and performance
- 29:34Conclusion and Q&A
related talks
26mFrom Spark to Elasticsearch and Back
41mInfrastructure for ML Applications
39mMachine Learning with Scala on Spark
38mReal Time ML Pipelines in Multi-Tenant Environments
24mUnder Pressure: Applying ML in Real Time
31mNLP text recommender system journey to automated training pipeline with Spark and Sagemaker
17mMachine Learning at the Speed of Light with Spark
23mRethinking scalable machine learning with Spark ecosystem