A tool to convert symbolic regression expressions into different formats.
A pandoc-like cli tool and library to convert symbolic regression expressions to convenient formats
pandoc-symreg: a pandoc-like tool to convert symbolic regression expressions to convenient formats.
Conversion tool for Symbolic Regression algorithms
Pandoc-Symreg is a Haskell library and CLI inspired by Pandoc for converting the output of Symbolic Regression tools to convenient formats for post analysis and documentation. It currently supports converting the output from
TIR(Transformation-Interaction-Rational Symbolic Regression)HL(HeuristicLab)Operon(Operon)Bingo(Bingo)GP-GOMEA(GP-GOMEA)PySR(PySR)SBP(SBP)EPLEX
And it can convert to
python(Numpy expression)math(Plain math expression)tikz(TikZ code to print a tree)latex(LaTeX equation)
This tool also supports changing floating-point numbers into parameter variables (named t or theta) and simplifying the expression using Equality Saturation as described in:
de França, Fabrício Olivetti and Kronberger, Gabriel. "Reducing Overparameterization of Symbolic Regression Models with Equality Saturation." Proceedings of the Genetic and Evolutionary Computation Conference. 2023. DOI: https://doi.org/10.1145/3583131.3590346
Installing
This tool can be installed via Cabal or Stack. The easiest way to install is via Haskell stack:
- Install Haskell Tool Stack
- Clone this repository
- Inside the project directory run the command
stack install
There are also binaries available at releases
Usage
Usage: pandoc-symreg (-f|--from ['tir'|'hl'|'operon'|'bingo'|'gomea'|'pysr'|'sbp'|'eplex'])
(-t|--to ['python'|'math'|'tikz'|'latex'])
[-i|--input INPUT] [-o|--output OUTPUT]
[-v|--varnames VARNAMES] [-p|--parameters] [--simplify]
Convert different symbolic expressions format to common formats.
Available options:
-f,--from ['tir'|'hl'|'operon'|'bingo'|'gomea'|'pysr'|'sbp'|'eplex']
Input expression format
-t,--to ['python'|'math'|'tikz'|'latex']
Output expression format
-i,--input INPUT Input file containing expressions. Empty string gets
expression from stdin. (default: "")
-o,--output OUTPUT Output file to store expressions. Empty string prints
expressions to stdout. (default: "")
-v,--varnames VARNAMES Comma separated list of variables names. Empty list
assumes the default of each algorithm (e.g,
"x,y,epsilon"). (default: "")
-p,--parameters Convert floating point numbers to free parameters.
--simplify Simplifies the expression using Equality Saturation.
-h,--help Show this help text
Contributing
If you want to add support to your SR algorithm, have a look at the file src/PandocSR.hs at the current parsers. You can either modify that file and make a Pull request or open an issue with the following informations:
- The name of your algorithm
- A list of supported univariate functions and their string representations
- A list of supported bivariate functions and their string representations
- A list of supported binary operators and their string representations. The string representation is sensitive to whether the operator is surrounded by space or not. See the source code for some examples.
Notice that we currently support a limited set of math functions and operators. See SRTree for the current list. Please open an issue describing any other function that you want to be supported.
If you want to add support to other output formats. Please open an issue with a description of the format and a link to the official project of the format, if any.
Bug reports and feature requests are welcome.
License
© 2023-2023 Fabricio Olivetti de Franca ([email protected]). Released under the GPL, version 3 or greater. This software carries no warranty of any kind. (See COPYRIGHT for full copyright and warranty notices.)