folivetti / srtree-opt

CLI to (re)optimize parameters of symbolic regression models

srtree-opt - a CLI tool to (re)optimize the numeric parameters of symbolic regression expressions

Optimize SR expressions using Levenberg-Marquardt.

Instalation

stack install

Command line help:

srtree-opt -h

Example usage:

Parse the Heuristic Lab expressions in test/feynman-III-17-37_expressions_heuristiclab.txt file using the training data in test/feynman_III_17_37_train.csv (with variable names in the header) taking the first $50$ rows as the training set and the next rows as validation, select the columns $0,1,2$ as the input variables and runs a maximum of $10$ iterations. It also simplifies the expressions using equality saturation prior to optimizing.

srtree-opt -f hl -i test/feynman-III-17-37_expressions_heuristiclab.txt -d test/feynman_III_17_37_train.csv -r 50 -c 0,1,2 --niter 10 --hasheader --simplify

Same as above but using column $4$ as the target variable.

srtree-opt -f hl -i test/feynman-III-17-37_expressions_heuristiclab.txt -d test/feynman_III_17_37_train.csv -r 50 -c 0,1,2 -t 4 --niter 10 --hasheader --simplify

By default the fitted expressions will be printed in stdout and the stats (sse of training and validation prior to optimization and after optimization) will be printed to stderr. You can specify the output files with the -o, -s flags.