Visual analytics (VA) systems with semantic interaction help users craft machine learning (ML) based solutions in various domains such as bio-informatics, finance, sports, etc.
However current semantic interaction based approaches are data and task-specific which might not generalize across different problem scenarios. In this project, we describe a novel technique of abstracting user intents and goals in the form of an interactive objective function which can guide any auto-ML based model optimizer (such as Hyperopt, Sigopt, etc.) to construct classification models catering to the expectations of the user. The objective function enables the auto-ML model optimizer to find the best classification models based on criteria's specified by the user. We believe abstraction of user intents in a mathematical form facilitates the generation of personalized ML solutions for any dataset, task or problem scenario.
