Bivariate Scientific Function Finding in a Sampled, Real-Data Testbed

摘要

This article reports the results of a study of domain-independent function finding based on a collection of several hundred real scientific data sets. Prior studies have introduced the controversial idea of discovering functional relatonships of interest to scientists directly from the data they collect. The evidence presented here supports the view that this is sometimes possible, but it also suggests how often purely data-driven discovery is not possible and how much more difficult it may be than has often been assumed. Experience with sampled examples of real scientific data suggests as well that emphasis on search in prior studies may have been misplaced. For the function-finding problems studied here, scientists typically propose only a handful of different functional relationships. The difficulty is not in searching through a large space of relationships but in evaluating a few common ones to determine if they are likely to be of scientific interest.