This repo contains the dataset generator library for creating symbolic regression datasets presented in the workshop paper "A Seq2Seq approach to Symbolic Regression" at NIPS 2020
Two types of datasets are available:
- Univariate datasets with combinations of polynomial and composition fuctions, with a fixed set of domain points (Used for our publication).
- Multivariate datasets of polynomial functions.
We try to reduce to a minimum ambiguities in the dataset, by introducing two key features:
- There is one to one correspondence between mathematical expression and realization set. (i.e. if sin(x)*cos(x) is in the dataset, cos(x)*sin(x) would be not)
- Expressions are always in a precise order
Create a python virtual enviroment with python -m venv env and activate it with source env/bin/activate
Install the library with pip install ./lib/ or pip install -e ./lib/ if you want the editable mode.
Eventually, run py.test test/ for checking that everything went well
Then check the two jupyters in the jupyter folder