EGG-SR: Embedding Symbolic Equivalence into Symbolic Regression via Equality Graph

1. Prerequisite of using these methods

• install the dependency package

pip install -r requirements.txt

• src/grammar: data structure for symbolic expressions. Before you run our program, you need to install the dataoracle by running

cd src/equality-graph/
pip install -e .

• src/equality-graph: our implementaiton of e-graph. Before you run our program, you need to install the dataoracle by running

cd src/equality-graph/
pip install -e .

2. Directory

Datasets

• datasets: the generated dataset. Every file represents a ground-truth expression.
• datasets/scibench: the data oracle API to draw data. Before you run our program, you need to install the dataoracle by running

cd datasets/scibench
pip install -e .

Methods

• algorithms/MCTS: the proposed method.
• algorithms/ProGED: from https://github.com/brencej/ProGED.
• algorithms/SPL: symbolic physics learner, from https://github.com/isds-neu/SymbolicPhysicsLearner.
• algorithms/E2E: End to end transformer for symbolic regression, from https://github.com/facebookresearch/symbolicregression.
• algorithms/odeformer: ODEFormer: Symbolic Regression of Dynamical Systems with Transformers.

Extra

• plots: the jupyter notebook to generate our figure.
• result: contains all the output of all the programs, the training logs.

3. Look at the summarized result

The experimental results are summarized in the result and plots folders.

Reference

• jitcode: https://github.com/neurophysik/jitcode

• DSRPytorch: https://github.com/dandip/DSRPytorch

• pytorch-minimize: https://github.com/rfeinman/pytorch-minimize