Touca Blog - How to track changes in the behavior of machine learning models using Touca
Like any other software component, machine learning models evolve over time. Data teams frequently make changes to their models to adjust and improve their effectiveness in meeting business needs. Understanding the true impact of these changes is essential but non-trivial.
Software engineering teams rely on a variety of testing tools and techniques to continuously validate their code changes. But these tools are not equally accessible or relevant to data teams. The common reasoning is that describing the expected outcome of machine learning models is more difficult than that of other software workflows. While this is true, part of the difficulty stems from trying to describe the behavior of these models in such ways to fit the requirements of established testing techniques like integration testing.
Touca, as an open-source regression testing system, is well positioned to address this difficulty and to effectively solve the testing needs of most data teams. This article showcases how Touca can help identify and evaluate changes in the behavior of a simple classification algorithm.
