notes_for_unsupervised_graph_classification.txt

Regarding unsupervised graph classification:
	
QGNN$ python train_graph_UnSup.py --dataset COLLAB --batch_size 4 --hidden_size 256 --num_epochs 100 --num_GNN_layers 4 --learning_rate 0.00005 --model_name COLLAB_bs4_hs256_fold0_k4_3

QGNN$ python train_graph_UnSup.py --dataset DD --batch_size 4 --hidden_size 256 --num_epochs 100 --num_GNN_layers 2 --learning_rate 0.001 --model_name DD_bs4_hs256_fold0_k2_0

QGNN$ python train_graph_UnSup.py --dataset IMDBBINARY --batch_size 4 --hidden_size 256 --num_epochs 100 --num_GNN_layers 2 --learning_rate 0.0005 --model_name IMDBBINARY_bs4_hs256_fold0_k2_1

See https://github.com/daiquocnguyen/Graph-Transformer for more details about unsupervised learning. Regarding the Pytorch implementation for the unsupervised learning, you should have Cython 0.29.13 and Scikit-learn 0.21 and then change to the `log_uniform` directory to perform `make` to build `SampledSoftmax`, and then add the `log_uniform` directory to your PYTHONPATH.