ML model from an idea to production with the help...
summary
Python pipeline moves a video‑ad completion idea into production using pandas, NumPy, scikit‑learn and TensorFlow with an in‑house Sroka data connector. A logistic regression is first implemented directly in page code then a decision tree is scikit‑learn → JavaScript port tested but causes latency, so production uses an HTTP Python service loading pickled models and serving predictions. Model separates traffic groups by ~10 percentage points and yields a ~4.5 point net lift; service health is tracked with Grafana and Mode monitoring.
AI-written summary · about
chapters (9)
- 0:00Data modeling and ML
- 2:37Fandom’s data ecosystem
- 6:50Ad completion ML project
- 7:54Data prep for ML
- 11:49Building ML models
- 16:49Communicating ML results
- 19:25Deploying Python models
- 21:36Monitoring models in production
- 24:14Key takeaways
related talks
37mAutomated Machine Learning: Building a Pipeline
30mOperationalizing Machine Learning
35mGetting Machine Learning Models Ready For Production Using Python
50mClosing the Machine Learning Loop using Python
25mContinuous ML Applications in Production
33mProduction ML Models in the Mobile Gaming Industry
32mMachine Learning Infrastructure at Stripe: Bridging from Python JVM
44mML in Production