SAGe

Abstract

Some of the most recent applications and services revolve around the analysis of time-series, which generally exhibits chaotic characteristics. This behavior brought back the necessity to simplify their representation to discover meaningful patterns and extract information efficiently. Furthermore, recent trends show how computation is moving back from the Cloud to the Edge of network, meaning that algorithms should be compatible with low-power embedded devices. A family of methods called Symbolic Analysis (SA) tries to solve this issue, reducing the dimensionality of the original data in a set of symbolic words and providing distance metrics for the obtained symbols. However, SA is usually implemented using application-specific tools, which are not easily adaptable, or mathematical environments (e.g. R, Julia) that do not ensure portability, or that require additional work to maximize computing performance. We propose here SAGe: a code generation tool that helps the user to prototype efficient and portable code, starting from a high-level representation of SA requirements. Other than exploiting similarities between SA pipelines, SAGe employs general code templates to build and deploy the code on different architectures, such as embedded devices, microcontrollers, and FPGAs. Preliminary results show a speedup up to 223x against Python implementations running on an x86 desktop machine and a notable increase in computational efficiency on a reconfigurable device.