Applying machine learning to software development to reduce bugs

on YouTube ↗38m

" summary

Claims that machine learning can locate concentrated risk in codebases and produce file-level risk scores using example Apache data. Demo builds a data warehouse and fact tables, trains a random forest with scikit-learn in Jupyter notebooks to flag buggy commits and produce a quad “brittle/unstable/stable/done” model. Surprising empirical result: roughly 3–5% of files hold most severe bugs while ~80% show none. Visualizations and warnings run on history and give actionable hotspots that the demo shows working.

AI-written summary · about

" chapters (8)

1. 0:00Introduction to machine learning in software development
2. 5:18Craftsmanship to industrial production
3. 10:59Predictive models for bug detection
4. 16:01Unified data model for engineering
5. 22:41Risk visualization with Numeric Quad Model
6. 25:29Live demo: Risk visualization on Apache HBase
7. 31:44Key takeaways for better engineering data
8. 33:50Q&A on bug severity and data

" related talks

• 53mBug prediction based on your code historyAlexey Tokar · SeleniumCamp 2018
• 48mUsing Machine Learning to find Linux bugsStef Walter · DevConf.CZ 2019
• 35mBetter C++ using Machine Learning on Large ProjectsNicolas Fleury, Mathieu Nayrolles · CppCon 2018
• 47mBug prediction based on your code historyAlexey Tokar · JavaDay Ukraine 2017
• 27mMachine Learning Practical ExamplePrasanth Anbalagan, Mark Yarborough · DevConf.CZ 2019
• 45m"Breaking black-box machine learning"Evelina Gabasova · Build Stuff Spain 2019
• 07mInterviewFrancesc Campoy Flores · Scale By The Bay 2018
• 32mReliable Machine LearningDavid Andrzejewski · Scale By The Bay 2019