Distributed Black-Box Model Explanation with Ray
summary
Parallelizes Kernel SHAP via Alibi across Ray actors to explain many independent predictions, packaging the explainer as a distributed actor pool and running it on a Kubernetes cluster. Uses Ray Serve for an HTTP endpoint and simple batching. Benchmarks show about 14× speedup on a 32‑vCPU machine and up to ~30× speedup when run on a 56‑vCPU cluster, turning half‑hour runs into ~1–2 minute jobs and enabling dataset‑level bias and feature‑effect analysis.
AI-written summary · about
chapters (8)
- 0:00Introduction to model explain
- 1:25Understanding black box models
- 2:51Methods for model explanation
- 4:46Complexity of SHAP values
- 7:48Speeding up tasks with Ray
- 13:31Deploying on a cluster
- 17:29Scaling with Ray Serve
- 21:59Insights and conclusion
related talks
15mRay: A System for Distributed Applications
01h 31mIntroduction to Ray for distributed and machine learning applications in Python
26mIntroduction to Distributed ML Workloads with Ray on Kubernetes
26mData Processing on Ray
20mThe Quick Journey to Using Ray: How We Implement Ray and Anyscale to Speed up our ML Processes
14mBuilding High Availability and Scalability Online Computing Applications on Ray
11mKeynote Q&A
23mImproving Ray for Large-scale Applications