This repository contains code for the paper Precision Machine Learning
, by Eric J. Michaud, Ziming Liu, and Max Tegmark.
Within notebooks/create-paper-figures
you can find the code which generated all the figures in the paper. There are some other jupyter notebooks within notebooks
to replicate other experiments mentioned in the paper which do not have a dedicated figure.
Many experiments involved running a grid search over things like network width, target equation, seed, activation function, etc. Configs defining these grid searches can be found in seml-experiments
. We used the seml package for executing these experiments, which call scripts (which are built with sacred) in the scripts
directory.
equations.csv
is a modified version of FeynmanEquations.csv
from Tegmark's Feynman Symbolic Regression Database, with some fixes and added equations.