code/gnn_1/README.md

Training GNN1

Training DiSCoMaT GNN1

for seed in 0 1 2; do
    bash run_discomat.sh $seed
done

Training DiSCoMaT w/o features GNN1

for seed in 0 1 2; do
    bash run_discomat_wo_features.sh $seed
done

Training DiSCoMaT w/o constraints GNN1

for seed in 0 1 2; do
    bash run_discomat_wo_constraints.sh $seed
done

Training v-DiSCoMaT GNN1

for seed in 0 1 2; do
    bash run_vdiscomat.sh $seed
done

All the above experiments can be parallelized easily. Hyper-parameters can be modified from within .sh scripts.

Generating results using different material ID thresholds

Once a model is trained, various result files using different MID thresholds can be generated by running the following script:

bash generate_res_files_for_all_alphas.sh

After generating the result files, refer to the evaluation folder to obtain compiled scores.