Data for the Paper "Projecting Latent RL Actions: Towards Generalizable and Scalable Graph Combinatorial Optimization"

This data repository contains the environments, logs, models, and analysis outputs used in the experiments of the associated paper. It is organized into four main experiment groups: action_time (inference/action selection time analysis), generalization (evaluation on unseen environments), hyperopt (Optuna-based hyperparameter optimization), and action_space_plots (UMAP visualization of latent action spaces). A shared config directory provides all environment generation and experiment configurations used across runs.

Each experiment follows a consistent structure with three main components: envs (benchmark environments and splits), logs (training/evaluation runs, checkpoints, and metrics), and optionally gae (Graph Autoencoder models, embeddings, and training logs).

A README file precisely describes the structure of the attached repository.

The data repository will be de-anonymized upon paper acceptance.