How to Prevent Algorithm-Porn
summary
Algorithm‑porn means adding complexity for ego not outcome and it kills projects through time and cost overruns and churn; a concrete win is stepping down from tf‑idf to a scikit‑learn count vectorizer and character tokenization which raises accuracy from about 80% to 90–93% and avoids deep learning. Recommendations: start with baseline and linear models, use pre‑trained nets like YOLO for vision, prefer hosted APIs and AWS Elastic Beanstalk or containers for deployment, adopt cookiecutter templates and checklists. A production HS‑code text classifier is built and deployed on AWS for high volumes.
AI-written summary · about
chapters (5)
- 0:00CHANGE WORLD THROUGH DATA
- 4:00Why is it happening?
- 6:01What can we really do about it?
- 7:381. Domain understanding 2. Problem understanding 3. Scope definition
- 10:581. Use domain knowledge and ready-made algorithms for feature selection and
related talks
41mBeyond the Algorithm: What Makes Machine Learning Work?
35mIntelligent Ruby: Getting Started with Machine Learning
30mMachine Learning: Beyond the Tutorial
06mInterview
36mApplied Machine Learning Playbook: From Idea to Production
19mDoing Machine Learning Without a Data Scientist
35mLevel Up Your Machine Learning Lifecycle
42mHow to Build Smart Applications Without a Machine Learning Background