We present P6, a declarative language for building high performance visualanalytics systems through its support for specifying and integrating machinelearning and interactive visualization methods. As data analysis methods basedon machine learning and artificial intelligence continue to advance, a visualanalytics solution can leverage these methods for better exploiting large andcomplex data. However, integrating machine learning methods with interactivevisual analysis is challenging. Existing declarative programming libraries andtoolkits for visualization lack support for coupling machine learning methods.By providing a declarative language for visual analytics, P6 can empower moredevelopers to create visual analytics applications that combine machinelearning and visualization methods for data analysis and problem solving.Through a variety of example applications, we demonstrate P6's capabilities andshow the benefits of using declarative specifications to build visual analyticssystems. We also identify and discuss the research opportunities and challengesfor declarative visual analytics.