Learning a Formula of Interpretability to Learn Interpretable Formulas
In this repository you can find the code and the supplementary material for our article about learning interpretable formulas with Genetic Programming.
You can find:
- the supplementary material for the article
- the datasets used for assessing our proposal
- the formulas used for the survey, both for decomposability and simulatability
- the results of the survey
- the code used for analyzing the results of the survey, extract the features and compute the linear model
- the results of the experiments with pyNSGP
For the code of GP part, you can refer to this repository.
How to reproduce the experiments
In order to replicate our experiment, you can use the run.py script. Such a script requires the datasets folder and takes as input:
- the name of the dataset
- the number of runs
- (otpional) the model to use for interpretability. If you specify an 'm' the script execute experiments with out model, otherwise with the nodes count based model.
If you use our code, please support our research by citing the related paper:
M. Virgolin, A. De Lorenzo, E. Medvet, F. Randone. "Learning a Formula of Interpretability to Learn Interpretable Formulas". arXiv preprint arXiv:2004.11170 (2020)