DySymNet 0.2.0

This package contains the official Pytorch implementation for the paper "A Neural-Guided Dynamic Symbolic Network for Exploring Mathematical Expressions from Data" accepted by ICML'24.

DySymNet

This repository contains the official Pytorch implementation for the paper A Neural-Guided Dynamic Symbolic Network for Exploring Mathematical Expressions from Data accepted by ICML'24.

🔥 Highlights

• DySymNet is a new search paradigm for symbolic regression (SR) that searches the symbolic network with various architectures instead of searching expressions in the large functional space.
• DySymNet possesses promising capabilities in solving high-dimensional problems and optimizing coefficients, which are lacking in current SR methods.
• DySymNet outperforms state-of-the-art baselines across various SR standard benchmark datasets and the well-known SRBench with more variables.

📦 Requirements

Install the conda environment and packages:

conda env create -f environment.yml
conda activate dysymnet

The packages have been tested on Linux.

📋 Getting started

The main running script is SymbolicRegression.py and it relies on configuring runs via params.py. The params.py includes various hyperparameters of the controller RNN and the symbolic network. You can configure the following hyperparameters as required:

parameters for symbolic network structure

Parameters	Description	Example Values
funcs_avail	Operator library	See params.py
n_layers	Range of symbolic network layers	[2, 3, 4, 5]
num_func_layer	Range of the number of neurons per layer of a symbolic network	[2, 3, 4, 5, 6]

Note: You can add the additional operators in the functions.py by referring to existing operators and place them inside funcs_avail if you want to use them.

parameters for controller RNN

Parameters	Description	Example Values
num_epochs	epochs for sampling	500
batch_size	Size for a batch sampling	10
optimizer	Optimizer for training RNN	Adam
hidden_size	Hidden dim. of RNN layer	32
embedding_size	Embedding dim.	16
learning_rate1	Learning rate for training RNN	0.0006
risk_seeking	using risk seeking policy gradient or not	True
risk_factor	Risk factor	0.5
entropy_weight	Entropy weight	0.005
reward_type	Loss type for computing reward	mse

parameters for symbolic network training

Parameters	Description	Example Values
learning_rate2	Learning rate for training symbolic network	0.01
reg_weight	Regularizaiton weight	5e-3
threshold	Prunning threshold	0.05
trials	Training trials for training symbolic network	1
n_epochs1	Epochs for the first training stage	10001
n_epochs2	Epochs for the second training stage	10001
summary_step	Summary for every n training steps	1000
clip_grad	Using adaptive gradient clipping or not	True
max_norm	Norm threshold for gradient clipping	1.0
window_size	Window size for adaptive gradient clipping	50
refine_constants	Refining constants or not	True
n_restarts	Number of restarts for BFGS optimization	1
add_bias	adding bias or not	False
verbose	Print training process or not	True
use_gpu	Using cuda or not	False
plot_reward	Plot reward curve or not	False

Note: threshold controls the complexity of the final expression, and is a trade-off between complexity and precision, which you can customise according to your actual requirements.

parameters for genearting input data

Parameters	Description	Example Values
N_TRAIN	Size of input data	100
N_VAL	Size of validation dataset	100
NOISE	Standard deviation of noise for input data	0
DOMAIN	Domain of input data	(-1, 1)
N_TEST	Size of test dataset	100
DOMAIN_TEST	Domain of test dataset	(-1, 1)

Additional parameters

results_dir configures the save path for all results

🤖 Symbolic Regression

We provide two ways to perform symbolic regression tasks.

Option1: Input ground truth expression

When you want to discover an expression for which the ground truth is known, for example to test a standard benchmark, you can edit the script SymbolicRegression.py as follows:

# SymbolicRegression.py
params = Params()  # configuration for a specific task
ground_truth_eq = "x_1 + x_2"  # variable names should be written as x_i, where i>=1.
eq_name = "x_1+x_2"
SR = SymbolicRegression(config=params, func=ground_truth_eq, fun_name=eq_name)  # A new folder named "func_name" will be created to store the result files.
eq, R2, error, relative_error = SR.solve_environment()  # return results

In this way, the function generate_data is used to automatically generate the corresponding data set $\mathcal{D}(X, y)$ for inference, instead of you generating the data yourself.

Then, you can run SymbolicRegression.py directly, or you can run it in the terminal as follows:

python SymbolicRegression.py

After running this script, the results will be stored in path ./results/test/func_name.

Option2: Load the data file

When you only have observed data and do not know the ground truth, you can perform symbolic regression by entering the path to the csv data file:

# SymbolicRegression.py
params = Params()  # configuration for a specific task
data_path = './data/Nguyen-1.csv'  # data file should be in csv format
SR = SymbolicRegression(config=params, func_name='Nguyen-1', data_path=data_path)  # you can rename the func_name as any other you want.
eq, R2, error, relative_error = SR.solve_environment()  # return results

Note: the data file should contains ($X_{dim} + 1$) colums, which $X_{dim}$ is the number of independent variables and the last colum is the corresponding $y$ values.

Then, you can run SymbolicRegression.py directly, or you can run it in the terminal as follows:

python SymbolicRegression.py

After running this script, the results will be stored in path ./results/test/func_name.

Output

Once the script stops early or finishes running, you will get the following output:

Expression: x_1 + x_2
R2: 1.0
error: 4.3591795754679974e-13
relative_error:  2.036015757767018e-06
log(1 + MSE):  4.3587355946774144e-13

🔗 Citing this work

If you find our work and this codebase helpful, please consider starring this repo 🌟 and cite:

@inproceedings{
li2024a,
title={A Neural-Guided Dynamic Symbolic Network for Exploring Mathematical Expressions from Data},
author={Wenqiang Li and Weijun Li and Lina Yu and Min Wu and Linjun Sun and Jingyi Liu and Yanjie Li and Shu Wei and Deng Yusong and Meilan Hao},
booktitle={Forty-first International Conference on Machine Learning},
year={2024},
url={https://openreview.net/forum?id=IejxxE9DO2}
}