Transformer-based Representation Learning for Optimization Problem Classification
Gjorgjina Cenikj, Gašper Petelin, Tome Eftimov · Zenodo (CERN European Organization for Nuclear Research) · 2023
This repository contains the data for calculating transformer-based representations for optimization problems. The files scaled_lhs_samples_dimension_X_Y_samples.csv contain the samples of the optimization problems where the objective function values are scaled, which are used for training the transformer models. The files lhs_samples_dimension_X_Y_samples.csv contain the samples of the optimization problems where the objective function values are not scaled, which are used for calculating the ELA features. The files ela_features_dimension_X_Y_samples.csv contain the calculated ELA features. The results_problem_classification.zip directory contains the results presented in the paper. The subfolder results_aggregations contains the results from the analysis of the predictive power of different aggregation functions The subfolder results_configurations contains the results from the analysis of the model parameters The subfolder results_downstream contains the results obtained with the best model parameters and all aggregation functions The subfolder results_ela contains the results from the problem classification using the ELA features