Encapsulating measurement uncertainty. UPROP: A Python library for Uncertainty Propagation

Abstract

In this work we present the UPROP Python library. A library for Uncertainty PROPagation, that allows developers to describe, store, and execute arithmetic operations on uncertain data. We overload arithmetic and comparison operators in order to achieve propagation of the input uncertainty to the output. To benchmark the library we develop real life applications with data derived from sensors. The main goal of the UPROP Library is to provide non-expert programmers a basic, user-friendly interface with a wide range of operations, which allows them to reason about uncertainty without ignoring it. The main advantage of the Uncertain type originates from its flexibility and minimalism. We develop GPS and sound recognition applications, using the UPROP library and uncertain sensor measurements. The usage of the library shows improved expressiveness and accuracy over the conventional usage of particle (one-sample) calculations. This approach leads to an average accuracy improvement, ranging from 1.2× to 40× with an average value of 7×. We further test the library using micro-benchmark comparisons with the state-of-the-art and list their differences. While the implementation of the UPROP library ensures the accessibility and expressivity of the Uncertain type it also has the potential to make a practical implementation impossible. The representation size embodies the classic speed-accuracy trade-off. Too high and the Uncertain type will be too slow for practical use; too low and it will be too inaccurate to solve real problems.