Production ML Models in the Mobile Gaming Industry
summary
Wildlife Studios describes taking a reinforcement‑learning auction model from research to production in six weeks, serving real‑time ad bids at millions of requests/sec and ~2 TB/day, training on billions of records. The pipeline uses Databricks + Spark for training, MLflow for model packaging and registry, GitLab CI/CD with Docker and Kubernetes/Flux for automated deployment (change model ID in repo to build/push/serve), Jaeger for tracing, and AWS infrastructure.
AI-written summary · about
chapters (7)
- 0:00Start
- 4:33Waterfall Optimization Problem
- 6:22Data Engineering Team (BA)
- 7:54Requirements
- 8:33Our methodology
- 10:11Machine Learning Project
- 21:23Waterfall Optimization Model Serving
related talks
25mContinuous ML Applications in Production
23mCustom-made games with machine learning and big data
32mMachine Learning for Game Developers
50mThe tech behind modern AI/ML use cases in Gaming
24mEngaging new and diverse audiences with ML
01h 05mMachine Learning Models and Productionization
26mManaging ML Models @ Scale
25mML model from an idea to production with the help...