Redhyte: Towards a self-diagnosing, self-correcting, and helpful analytic platform

Abstract

We present a platform named Redhyte, short for an interactive platform for “Rapid exploration of data and hypothesis testing”. Redhyte aims to augment the conventional statistical hypothesis testing framework with data-mining techniques in a bid for more wholesome and efficient hypothesis testing. The platform is self-diagnosing (it can detect whether the user is doing a valid statistical test), self-correcting (it can propose and make corrections to the user’s statistical test), and helpful (it can search for promising or interesting hypotheses related to the initial user-specified hypothesis). In Redhyte, hypothesis mining consists of several steps: context mining, mined-hypothesis formulation, mined-hypothesis scoring on interestingness, and statistical adjustments. To capture and evaluate specific aspects of interestingness, we developed and implemented various hypothesis-mining metrics. Redhyte is an R shiny web application and can be found online at https://tohweizhong. shinyapps.io/redhyte, and the source codes are housed in a GitHub repository at https://github.com/tohweizhong/redhyte.