Fast feedback on every change you do to your software codebase is essential for a rapid and safe development cycle. Failed test runs need attention from human experts that can quickly analyze and take necessary action. Our problem: the number of failed tests results is higher than our little group of domain experts can handle. We need a more efficient and effective way to remove the risks of having un-scrubbed failed test results lying around. We developed a machine learning tool based on test results clustering and this allows us to open up for even more automation for handling test results.
This presentation is about what we needed and how we solved it with only an entry-level machine-learning course and some Python skills. Clustering techniques can be useful in any field and the path from an idea to get something of value running can be short with the right approach and tools.