Trained Model Weights and Evaluation Data for Automated Architectural Layout Generation using Reinforcement Learning

This dataset contains the replication assets for the paper titled "Automated architectural layout generation using Reinforcement Learning with a Curriculum Learning strategy". It contains the following files:1. trained_models.zip: - best_model.zip: The neural network weights for the final Stage 4 tuned CL-PPO agent. - best_model_vec_normalize.pkl: The running mean and variance normalizer for observations. 2. evaluation_data.zip: - comparative_evaluation_results.json: Raw simulation output log for 1,000 evaluation episodes comparing CL-PPO with baseline searches (MC-Global and MC-Local). - ablation_evaluation_results.json: Raw output log for the ablation studies evaluating Stage 2 and Stage 3 performance metrics. - *.monitor.csv: Raw monitor logs containing step-by-step reward, episode length, and exact training duration. The companion source code for the environment and evaluation pipeline is open-sourced on GitHub at: https://github.com/rrucki/generative-architectural-design