A Coding Translation to Increase the Efficiency of Programmatic Data Analyses

Programmatic data analysis (PDA) is the design of software that is capable of performing a range of data analyses. When developed well, the computer programming code can enhance the generalizability and efficiency of data analyses, along with facilitating dynamic calculations in realtime applications. Programmatic designs may introduce some trade-offs, such as making the data analysis software more flexible at the expense of increased running time in calculations. This study identifies a class of programmatic designs in the R language that significantly degrade the run-time performance of data analyses relative to more direct coding syntax. To address this issue, this study introduces a translation algorithm of programmatic coding statements for use in the R language. These translations are based upon natural language processing of the coding statement and metaprogramming techniques for constructing new statements. The study's experiments demonstrate that utilizing the translated coding statement improves the running time performance. Reductions in the running time up to 60% are observed, with additional gains for increasing sample sizes of data. The proposed translation algorithm resolves the observed trade-offs of using programmatic data analyses for this class of problems, showing comparable performance with direct coding statements that lack programmatic designs. These promising results suggest that coding translations can be a valuable tool for improving the running time performance of programmatic data analyses.