An End-to-End Machine Learning Ecosystem in a Quarter
summary
Claims a working end-to-end pipeline built in roughly a quarter that trains and serves per-account models at scale: uses pandas, scikit-learn, flask and hadoop to extract features, run randomized cross‑validation over a huge parameter space, and parallelize training across Hadoop nodes. It outputs pickled models saved to HDFS, served by a cached Flask API called Crystal Ball, and a wrapper framework selfie automates searches; AUROC ranges up to 0.99 and typically sits around 0.8, models retrain regularly and compare to previous versions before deployment.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 6:15Predicting Customer Satisfaction
- 7:00Data Science
- 8:55Machine Learning
- 14:50Automatic Tuning
- 18:50Building Models
- 21:05Making Models Accessible
- 25:25Questions
- 33:46Handling different types of tickets
related talks
53mEnd-to-End ML
37mAutomated Machine Learning: Building a Pipeline
39mEnd to End Machine learning pipelines for Python driven organizations
30mOperationalizing Machine Learning
37mSoftware Development for Machine Learning in Python
38mBuilding robust machine learning systems
50mDemocratizing AI - Back-fitting end-to-end machine learning
46mThe Python Data Ecosystem: Navigating a fragmented landscape