What Users Don't Expect about Exploratory Data Analysis on Approximate Query Processing Systems

Pangloss implements "Optimistic Visualization", a method that gives analysts confidence to use approximate results for exploratory data analysis. In this paper, we outline how analysts' experience with an approximate visualization system did not match their intuitions. These observations have implications for the design of future data exploration systems that expose uncertainty. We also describe requirements for approximate query engines to enable the next generation of exploratory visualization systems.