ogbn_proteins

ogbn-proteins

We simply apply a random partition to generate batches for both mini-batch training and test. We set the number of partitions to be 10 for training and 5 for test, and we set the batch size to 1 subgraph. We initialize the features of nodes through aggregating the features of their connected edges by a Sum (Add) aggregation.

Default

--use_gpu False 
--cluster_number 10 
--valid_cluster_number 5 
--aggr add 	#options: [mean, max, add]
--block plain 	#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
--use_one_hot_encoding False
--mlp_layers 2
--norm layer
--hidden_channels 64
--epochs 1000
--lr 0.001
--dropout 0.0
--num_evals 1

DyResGEN-112

Train the model that performs best

python main.py --use_gpu --conv_encode_edge --use_one_hot_encoding --num_layers 112 --block res+ --gcn_aggr softmax --t 1.0 --learn_t --dropout 0.1 

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

python test.py --use_gpu --conv_encode_edge --use_one_hot_encoding --num_layers 112 --block res+ --gcn_aggr softmax --t 1.0 --learn_t --dropout 0.1

Test by multiple evaluations (e.g. 5 times)

python test.py --use_gpu --num_evals 5 --conv_encode_edge --use_one_hot_encoding --num_layers 112 --block res+ --gcn_aggr softmax --t 1.0 --learn_t --dropout 0.1 

Train ResGCN-112

python main.py --use_gpu --conv_encode_edge --use_one_hot_encoding --num_layers 112 --block res --gcn_aggr max

Train with different GCN models with 28 layers on GPU

SoftMax aggregator with learnable t (initialized as 1.0)

python main.py --use_gpu --conv_encode_edge --use_one_hot_encoding --num_layers 28 --block res+ --gcn_aggr softmax --t 1.0 --learn_t

PowerMean aggregator with learnable p (initialized as 1.0)

python main.py --use_gpu --conv_encode_edge --use_one_hot_encoding --num_layers 28 --block res+ --gcn_aggr power --p 1.0 --learn_p

Apply MsgNorm (message normalization) layer (e.g. SoftMax aggregator with fixed t (e.g. 0.1))

Not learn parameter s (message scale)

python main.py --use_gpu --conv_encode_edge --use_one_hot_encoding --num_layers 28 --block res+ --gcn_aggr softmax_sg --t 0.1 --msg_norm

Learn parameter s (message scale)

python main.py --use_gpu --conv_encode_edge --use_one_hot_encoding --num_layers 28 --block res+ --gcn_aggr softmax_sg --t 0.1 --msg_norm --learn_msg_scale

ResGEN

SoftMax aggregator with fixed t (e.g. 0.001)

python main.py --use_gpu --conv_encode_edge --use_one_hot_encoding --num_layers 28 --block res+ --gcn_aggr softmax_sg --t 0.001

PowerMean aggregator with fixed p (e.g. 5.0)

python main.py --use_gpu --conv_encode_edge --use_one_hot_encoding --num_layers 28 --block res+ --gcn_aggr power --p 5.0

ResGCN+

python main.py --use_gpu --conv_encode_edge --use_one_hot_encoding --num_layers 28 --block res+ --gcn_aggr mean

ResGCN

python main.py --use_gpu --conv_encode_edge --use_one_hot_encoding --num_layers 28 --block res --gcn_aggr mean

PlainGCN

python main.py --use_gpu --conv_encode_edge --use_one_hot_encoding --num_layers 28 --gcn_aggr mean