Things to bench

This repository shows an example of how one can work with the symbolic regression benchmark cp3-bench. Furthermore, one can find our results related to our paper.

Benchmarking symbolic regression algorithms using cosmological data.

The authors of this paper would like to close by stating that we are open to cooperation in regards to develop the framework of cp3-bench further, including adding more algorithms to the suite, but also adding new features and improving compatibility to other platforms. We suggest contacting the authors per e-mail and code contributions done via pull requests.

Getting started

To try out this repository clone it with recursion:

git clone https://github.com/CP3-Origins/Things-to-bench.git --recursive

To run the benchmark go to the cp3-bench folder for installation instructions.

Notes on test datasets

Here you can see specifications of the test dataset found in datasets.

F1 with data range x=[-10,1] N=1000

Function: $x^5 - 2*x^3 + x$

F2 with data range x=[-10,10] N=1000

Function: $0.3xsin(2*\pi*x)$

F3 with data range x=[0,10] N=1000

Function: $x^3*exp(-x)cos(x)sin(x)(sin(x)^2cos(x)-1)$

F4 in interval x,y=[-3,3] N=100

Function: $2.5x^4 - 1.3x^3 + 0.5y^2 - 1.7y$

F5 in interval x,y=[-3,3] N=100

Function: $1.5e^x + 5.0cos(y)$

F6 in interval x,y=[0.3,4] N=100

Function: $e^{-(x-1)^2}/( 1.2+(y-2.5)^2)$

F7 in interval x,y,z=[-5,5] N=30

Function: $0.23+14.2*(x + y)/(3*z)$

F8 in interval x,y,z=[-5,5] N=30

Function: $6.78+11sqrt(7.23xyz)$

These equation are inspired by: https://arxiv.org/pdf/2211.10873.pdf

Citing this work

If you find cp3-bench and this example useful, please cite:

@misc{thing2024cp3bench,
      title={cp3-bench: A tool for benchmarking symbolic regression algorithms tested with cosmology}, 
      author={Mattias E. Thing and Sofie M. Koksbang},
      year={2024},
      eprint={2406.15531},
      archivePrefix={arXiv},
      primaryClass={astro-ph.IM},
      url={https://arxiv.org/abs/2406.15531}, 
}