Building an A.I. Cloudy Sky: What We Learned at PredictionIO
summary
PredictionIO is an open-source ML server (Scala/Java) that surfaces ten concrete platform decisions: event server for mixed batch/stream imports, three real examples (recommendations, churn, spam), tradeoffs between real-time scoring and batch retrain schedules, messy data ingestion and change logs, feature engineering choices, label construction, human-facing evaluation metrics, and automated monitoring. The project claims top‑five GitHub ML status, shows practical integrations and failures, and argues the next step is AI for AI to automate hyperparameter tuning, model selection and per-customer customization.
AI-written summary · about
chapters (8)
- 0:00Introduction and background
- 3:50Simplicity vs flexibility in AI platforms
- 13:49Defining prediction problems and business goals
- 21:11Data types and event server challenges
- 26:54Feature generation and model training
- 33:38Monitoring and maintaining models
- 35:26Business challenges in platform building
- 37:10Future of PredictionIO and Q&A
related talks
40mCloud-based Machine Learning Offerings - Beginner's Perspective
18mAI Owes its Rebirth to Open Source and the Cloud
31mCreating recommender systems with Spark, Scala and Prediction.Io
45mPredictive Computing: A.I. Gets Real
58mThe path between developing and serving machine learning models.
13mBuilding Scalable AI/ML Platforms for Industrial IoT
23mAI-Driven Enterprise Intelligence
44mMachine learning @ booking.com