Early Test Strategies in ML Model Development Phase
summary
Early QA participation locates overfitting, underfitting and data drift during model engineering rather than after validation, using staged training data and stop criteria to avoid degradation. Example: a real estate price regressor trained on bedroom and sqft attributes shows optimal accuracy then degrades when irrelevant features (color) are added. The workflow uses TensorFlow, scikit-learn and learning curve utilities to auto-detect fit problems, emit diagnostic graphs, then convert manual checks into repeatable scripts.
AI-written summary · about
chapters (6)
- 0:00Introduction to session
- 0:29What is machine learning
- 5:12Collaborating with developers
- 7:00Overfitting and underfitting
- 12:27Automating test detection
- 13:54Shift left testing importance
related talks
26mTips on Quality Assurance process in Machine Learning projects
30mHow to Properly Test ML Models & Data
31mTesting Machine Learning Models
30mDetecting drift: how to evaluate and explore data drift in machine learning systems
42mTesting a Data Science Model
42mMasterclass: The right path to ML
26mTesting a Data Science Model
53mUnleashing the Black Box: Testing Machine Learning Systems