Skip to content

Latest commit

 

History

History

ogbg_mol

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 
 
 
 
 

ogbg_mol

The code is shared by two molecular datasets: ogbg_molhiv and ogbg_molpcba.

Default

--use_gpu False 
--dataset ogbg-molhiv
--batch_size 32
--block res+	#options: [plain, res, res+]
--conv gen
--gcn_aggr max 	#options: [max, mean, add, softmax, softmax_sg, softmax_sum, power, power_sum]
--num_layers 3
--conv_encode_edge False
--add_virtual_node False
--mlp_layers 1
--norm batch
--hidden_channels 256
--epochs 300
--lr 0.01
--dropout 0.5
--graph_pooling mean  #options: [mean, max, sum]

ogbg_molhiv: DyResGEN

Train

python main.py --use_gpu --conv_encode_edge --num_layers 7 --dataset ogbg-molhiv --block res+ --gcn_aggr softmax --t 1.0 --learn_t --dropout 0.2 --lr 0.0001

Test (use pre-trained model, download from Google Drive)

python test.py --use_gpu --conv_encode_edge --num_layers 7 --dataset ogbg-molhiv --block res+ --gcn_aggr softmax --t 1.0 --learn_t

ogbg_molpcba: ResGEN + virtual nodes

Train

python main.py --use_gpu --conv_encode_edge --add_virtual_node --mlp_layers 2 --num_layers 14 --dataset ogbg-molpcba --block res+ --gcn_aggr softmax_sg --t 0.1

Test (use pre-trained model, download from Google Drive)

python test.py --use_gpu --conv_encode_edge --add_virtual_node --mlp_layers 2 --num_layers 14 --dataset ogbg-molpcba --block res+ --gcn_aggr softmax_sg --t 0.1 --model_load_path ogbg_molpcba_pretrained_model.pth