VisEvol: Visual Analytics to Support Hyperparameter Search through Evolutionary Optimization

During the training phase of machine learning (ML) models, it is usually necessary to configure several hyperparameters. This process computationally intensive and requires an extensive search infer best hyperparameter set for given problem. The challenge exacerbated by fact that most ML models are complex internally, involves trial-and-error processes could remarkably affect predictive result. Moreover, each algorithm potentially intertwined with others, changing might result in unforeseeable impacts on remaining Evolutionary optimization a promising method try address those issues. According this method, performant stored, while remainder improved through crossover mutation inspired genetic algorithms. We present VisEvol, visual analytics tool supports interactive exploration hyperparameters intervention evolutionary procedure. In summary, our proposed helps user generate new evolution eventually explore powerful combinations diverse regions space. outcome voting ensemble (with equal rights) boosts final performance. utility applicability VisEvol demonstrated two use cases interviews experts who evaluated effectiveness tool.