Ydo-W / SR-RFND

SR-RFND：Symbolic regression for redundant features and noise data

Description

This is the implementation of manuscript “***”.

System requirements

Installation

• Python 3.7+
• PtTorch 1.8.0+cu111
• PySR 0.12.0

Quick start

Examples

We provide in this repository the verification of SR-RFND on two known formulas, as follows:

1. $$E=\frac{q_1}{4\pi\epsilon r^2},$$
2. $$E=\frac{1}{2}m(v^2 + u^2 + w^2)$$

The demos for these two examples are stored under ./examples_of_known_formulas-1/ and examples_of_known_formulas-2 respectively.

Data

Users can use our datasets, stored under ./{example folder}/datasets/, or re-generate them by running ./{example folder}/dataset_made.py.

Running

Direct Symbolic Regression

1. Run ./{example folder}/directly_SR/SR.py.

Feature Filtering & Symbolic Regression

1. Run ./{example folder}/pipeline_on_noise_data/train_baseline.py to train the baseline network for performing multivariate regression task.
2. Run ./{example folder}/pipeline_on_noise_data/RFE_feature_selection.py to perform the feature filtering, the results will be reported in Feature importance.log under the same folder.
3. Run ./{example folder}/pipeline_on_noise_data/SR.py to perform symbolic regression and evaluation.

Sample Filtering & Feature Filtering & Symbolic Regression

1. Run ./{example folder}/train_baseline_ncr/train_baseline.py to train the baseline network with neighborhood consistency regularization. Then move the trained model file to ./{example folder}/pipeline_on_selected_data/checkpoints/.
2. Run ./{example folder}/pipeline_on_selected_data/data_selection.py to perform sample filtering.
3. Run ./{example folder}/pipeline_on_noise_data/train_baseline.py to train the baseline network for performing multivariate regression task.
4. Run ./{example folder}/pipeline_on_noise_data/RFE_feature_selection.py to perform the feature filtering, the results will be reported in Feature importance.log under the same folder.
5. Run ./{example folder}/pipeline_on_noise_data/SR.py to perform symbolic regression and evaluation.