Lightning Talks
summary
Argues that most “ML for testing” tools are marketing, with the main useful application being defect classification learned from historical failures, while defect prediction often amounts to simple log-pattern search. Describes a working pipeline using neural preprocessing of noisy stack traces, vector clustering, and the released report-portal rp4.0; plus a Lambda-based CI feedback loop for Gerrit/Jenkins and an early prototype of an object-as-test runner that exports test instances, supports dependency injection and filters, and targets Allure compatibility.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:10Machine learning in testing
- 3:37AI testing market analysis
- 5:27AI testing hype problems
- 8:59ML log classification
- 12:01Gerrit topic feedback fix
- 15:30Lambda event processing
- 24:54OOP test concept
- 25:57New testing approach
- 30:53Tool compatibility
- 31:56Conclusion
