viztrackr: Tracking and discovering plots via automatic semantic annotations
Data analyses often produce many different data visualizations. Keeping track of these plots is crucial for both correctness and reproducibility of analytic results. Analysts typically resort to direct use of filenames and paths to organize and label their plots. Unfortunately, such ad hoc approaches do not scale well to longer and more complex analyses. Furthermore, locating specific plots months or years after the fact, when the chosen naming scheme has likely been forgotten, can be time consuming and painful. We propose a system which automatically tracks visualizations and annotates them with meaningful, searchable metadata. Beyond the benefits to individual analysts, the ability to search through plots created by others to discover analyses relevant to a particular dataset or research question is a powerful tool for facilitating collaboration and advancing science within multi-analyst, multi-project research departments and the wider scientific community. We present the viztrackr framework, a tool for tracking, automatically annotating, discovering, and reproducing statistical plots created in the R statistical programming language.