L2Lift: a learning-based cost estimation method for CAD variable ordering

L2Lift====== L2Lift is a learning-based cost estimation method for CAD variable ordering. Install------- Prerequisites - NVIDIA GPU with compatible drivers (CUDA 11.8+ required) - Miniconda or Anaconda installed - Maple installed (See https://www.maplesoft.com/) Quick Setup Create Conda Environment conda env create -f env.yml conda activate cad Verify Installation import torch print(torch.__version__) print(torch.cuda.is_available()) Troubleshooting If environment creation fails: - Check NVIDIA drivers: nvidia-smi - Install pytorch from https://pytorch.org/get-started Train----- To train the root prediction model, run: python src/train/train_models.py To train the time prediction model, modify variable "label_file" and "model_save_path"in src/train/train_models.py. Models are saved in models/, logs in logs/. Predict------- example.sh provides a simple command example for L2Lift. If you want to executeGL2Lift, please add the --greedy option in the command. l2lift.py will not output any results to the terminal. All prediction results will bestored in the svos directory under each polynomial directory. Evaluate-------- To evaluate the prediction results, run: python src/evaluate.py Build groud truth----------------- If you need to obtain the Projection results and CAD results for polynomials, refer to: src/mpl_utils/readme.md