Note: built up based on https://github.com/Thinklab-SJTU/One-Shot-Cardinality-NN-Solver

We use the following packages

torch                  1.11.0
torch-geometric        2.1.0
torch-scatter          2.0.9
torch-sparse           0.6.13
torch-spline-conv      1.2.1

First, install the required packages

pip install torch==1.11.0+${CUDA} --extra-index-url https://download.pytorch.org/whl/${CUDA}
pip install torch-sparse==0.6.13 torch-spline-conv==1.2.1 torch-scatter==2.0.9 -f https://data.pyg.org/whl/torch-1.11.0+${CUDA}.html
pip install torch-geometric==2.1.0.post1
pip install ortools easydict matplotlib yfinance pandas cvxpylayers xlwt pyyaml gurobipy

We have the following methods

• CPU methods : random / gurobi / scip / greedy
• CardNN methods : cardnn-s / cardnn-gs / cardnn-hgs
• CardNN-noTTO methods : cardnn-notto-s / cardnn-notto-gs / cardnn-notto-hgs
• EGN method : egn-naive
• Ours (UCom2) : ucom2

Facility location

Below are the datasets for the facility location problem

• Datasets : rand500 / rand800 / starbucks / mcd / subway

Below are the commands for running CPU methods

• CPU methods : random / gurobi / scip / greedy

python facility_location_[method].py --cfg cfg/facility_location_[dataset].yaml --timestamp [seed]

For example, to run the method "random" on "rand500" datasets, you should run

python facility_location_random.py --cfg cfg/facility_location_rand500.yaml --timestamp 0

Below are the commands for running CardNN and CardNN-noTTO methods

• CardNN methods : cardnn-s / cardnn-gs / cardnn-hgs
• CardNN-noTTO methods : cardnn-notto-s / cardnn-notto-gs / cardnn-notto-hgs

python facility_location_[method].py --cfg cfg/facility_location_[dataset].yaml --lr [learning_rate] --reg [gumbel_sigma] --timestamp [seed]

For example, to run the method "cardnn-hgs" on "rand500" datasets, you should run

python facility_location_cardnn-hgs.py --cfg cfg/facility_location_rand500.yaml --lr 1e-4 --reg 0.25 --timestamp 0

Below are the commands for running EGN method and our method

• EGN method : egn-naive
• Ours (UCom2) : ucom2

python facility_location_[method].py --cfg cfg/facility_location_[dataset].yaml --lr [learning_rate] --reg [constraint_coefficient] --timestamp [seed]

For example, to run the proposed method "ucom2" on "rand500" datasets, you should run

python facility_location_ucom2.py --cfg cfg/facility_location_rand500.yaml --lr 1e-2 --reg 1e-2 --timestamp 0

For the RL method, we adapted code from https://github.com/ai4co/rl4co. Please follow the installation guidance thereof.

See the files in the RL folder: facility_location_rl_*.py.

For example, to run the RL method on "rand500" datasets, you should run

python RL/facility_location_rl_rand500.py --ds rand500_train

Maximum coverage

Below are the datasets for the maximum coverage problem

• Datasets : rand500 / rand1000 / twitch / rail

Below are the commands for running CPU methods

• CPU methods : random / gurobi / scip / greedy

python max_cover_[method].py --cfg cfg/max_cover_[dataset].yaml --timestamp [seed]

For example, to run the method "random" on "rand500" datasets, you should run

python max_cover_random.py --cfg cfg/max_cover_rand500.yaml --timestamp 0

Below are the command for running CardNN and CardNN-noTTO methods

• CardNN methods : cardnn-s / cardnn-gs / cardnn-hgs
• CardNN-noTTO methods : cardnn-notto-s / cardnn-notto-gs / cardnn-notto-hgs

python max_cover_[method].py --cfg cfg/max_cover_[dataset].yaml --lr [learning_rate] --reg [gumbel_sigma] --timestamp [seed]

For example, to run the method "cardnn-hgs" on "rand500" datasets, you should run

python max_cover_cardnn-hgs.py --cfg cfg/max_cover_rand500.yaml --lr 1e-5 --reg 0.15 --timestamp 0

Below are the command for running EGN method and our method

• EGN method : egn-naive
• ours (UCom2) : ucom2

python max_cover_[method].py --cfg cfg/max_cover_[dataset].yaml --lr [learning_rate] --reg [constraint_coefficient] --timestamp [seed]

For example, to run the proposed method "ucom2" on "rand500" datasets, you should run

python max_cover_ucom2.py --cfg cfg/max_cover_rand500.yaml --lr 1e-5 --reg 0.15 --timestamp 0

For the RL method, we adapted code from https://github.com/ai4co/rl4co. Please follow the installation guidance thereof.

See the files in the RL folder: max_cover_rl_*.py.

For example, to run the RL method on "rand500" datasets, you should run

python RL/max_cover_rl_rand500.py --ds rand500_train

Batch files

We also provide bash files for conveniently executing experiments for each method.

Below are the commands for running CPU methods

• CPU methods : random / gurobi / scip / greedy

bash [method].sh

For example, to execute experiments with the method "random", you should run

bash random.sh

For the other methods, two bash files are provided for two problems (facility location and maximum coverage), perspectively.

bash [fl/mc]_{method}.sh

For example, to execute experiments on the facility location problem with the proposed method "ucom2", you should run

bash fl_ours.sh