Methods for time-series analysis have been developed for diverse problems in areas ranging from astrophysics, biomedicine, finance, and industry. From this broad interdisciplinary toolbox, the choice of which analysis method to use for a given problem is typically subjective, leaving open the question of whether alternative methods could outperform those chosen by a given researcher. Here we describe the systematic, data-driven selection of methods for a particular time-series analysis problem through systematic comparison across a comprehensive library of over 7000 time-series features, using the hctsa toolbox. This highly comparative approach to leveraging interdisciplinary knowledge on time series has proven successful on a wide range of domain problems. Due to redundancy across the library, we are able to distill the interdisciplinary time-series analysis literature down to a canonical set of 22 features that exhibits high accuracy across 93 time-series classification tasks. This reduced feature-based representation is used to automatically organize diverse time-series data in an interactive living library, CompEngine (https://www.comp-engine.org/).