Getting Machine Learning Models Ready For Production Using Python
summary
Describes an eight-week production cycle that moves notebooks to a Docker-served model by splitting work into sprints: literature review, EDA (pandas, pandas-profiling), feature engineering, model development (scikit-learn, PyTorch, TensorFlow), then refactor to a predict wrapper template, unit tests and integration/deployment. Uses git, Docker, Airflow for automated training, and DVC for dataset versioning; sometimes hand-implements NumPy matrix code to meet a 100 ms latency constraint. Claims this process delivers production models on real traffic (≈1M tx/day).
AI-written summary · about
chapters (8)
- 0:00Introduction
- 2:24What typically happens
- 6:40Agile Production Cycle Plan
- 11:24Model Development
- 15:58Model Optimization
- 20:36In Testing
- 21:43Recap
- 22:30QA
related talks
32mDeploying Python models to production
35mHow to deploy machine learning models to production (frequently and safely)
01h 18mPutting Tensorflow Models into Production
32mAutomating Production Production Level Machine Learning on AWS
24mFrom Model to Production: Automating the ML Lifecycle with CI/CD
57mDevOps for Machine Learning
56mDevOps for Machine Learning
01h 05mMachine Learning Models and Productionization