Welcome to the Feynman Symbolic Regression Database!


As opposed to linear regression, where a dataset is fit to a linear function of the given input variables, symbolic regression tackles the much harder problem where the unknown function can correspond to any combination of mathematical symbols. To facilitate development and testing of algorithms for this, this database contains 120 symbolic regression mysteries as described in the paper AI Feynman: a Physics-Inspired Method for Symbolic Regression, Udrescu & Tegmark (2019), arXiv:1905.11481.

Each regression mystery corresponds to a table of numbers, whose rows are of the form {x1,x2,...,y}, where y = f(x1,x2,...); the challenge is to discover the correct symbolic expression for the mystery function f. If you're eager to get started, simply download and uncompress Feynman_with_units.tar.gz below and try dicovering 100 equations from the Feynman Lectures! If that's too easy, please try the 20 more challenging mysteries in bonus_without_units.tar.gz. You'll find the correct answers in the spreadsheets below. If you prefer the mysteries simplified by dimensional analysis, try the "without_units" files instead, you'll also want to download units.csv, the unit table described in the paper.
Downloadable files:
If you have questions, please contact Silviu at sudrescu@mit.edu