The files are not fully organized yet; the uploaded files are key components of the original files.

Once the original files are organized, they will be uploaded.

Installation and dependencies

pip install -r requirements.txt

We use Gurobi 11.0 as MIP solver. After activating the Gurobi license, the environment setup is complete.

Quick start guide

Note that the default value for core_num is 1, which can be extremely slow. It is recommended to set this parameter as high as possible.

Step1: generate initial state for dataset generation

python gen_instance_dataset/gen_instance.py --instance 80_8_std --core_num 1

setp2: generate high-quality,middle and low-quality commitments

python gen_instance_dataset/gen_contrastive_json.py --instance 80_8_std --core_num 1

step3: generate initial prediction and neighborhood prediction datasets

python gen_instance_dataset/gen_initial_prediction_dataset.py --instance 80_8_std --core_num 1
python gen_instance_dataset/gen_neighborhood_search_dataset.py --instance 80_8_std --core_num 1

step4: train two distinct GNN policies
open train_file\train_initial_prediction_3bin.ipynb set instance name and run all cells.
open train_file\train_neighborhood_search_3bin.ipynb set instance name and run all cells.
step5: evaluate on test set
open evaluate.ipynb and run all cells.