Deploying Machine Learning Models in Practice
summary
Compares deployment options for real systems, not just models, and shows concrete tooling: exports and runtimes for scikit‑learn, Spark ML and TensorFlow using PMML, PFA and ONNX plus MLeap and TF Serving; demonstrates a Scala library Aardvark that exports Spark ML pipelines to PFA export and experimental flow to TensorFlow graphs for TF Serving; argues containers (MAX, Seldon, Clipper) help but don’t solve versioning, pipeline skew, latency or monitoring and urges pipeline portability, CI and continuous retraining.
AI-written summary · about
chapters (10)
- 0:00Introduction to ML deployment
- 2:24The complex ML workflow
- 6:02What is model deployment
- 10:28Challenges in production
- 14:58Containerized solutions
- 18:59Standardized serving frameworks
- 23:08Open standards for formats
- 30:45Deep dive into PFA and Aardvark
- 38:54Monitoring and feedback loops
- 44:11Q&A and concluding thoughts
related talks
33mOpen Standards for Machine Learning Deployment
34mOpen Standards For Machine Learning Model Deployment
33mDeploying and Scaling Spark ML and Tensorflow AI Models
46mProduction Model Deployment
41mOpen standards for machine learning deployment
33mOpen Standards for Machine Learning Deployment
39mApplying deployment oriented mindset for building Machine Learning models
24mMaking Spark ML Models Portable Know Your Options