hyperparams.yml

batch_size: 1000                      # Batch size for graph sampling
epochs: 500                          # Number of epochs for training
test_epochs: 500                      # Number of epochs for testing
weight_decay: 0.00001                 # Weight decay for ...
warmup: 50                            # Number of wamup steps for ...
g_temperature: 1.0                    # Temperature of augmentation model g (?)
g_hidden: 64                          # Number of hidden units for augmentation model g (?)
g_dropout: 0.1                        # Dropout for encoder for augmentation model g
g_nlayer: 1                           # Number of layers for encoder for augmentation model g
mlpx_dropout: 0.1                     # Dropout for MLPX model in augmentation model g
f_hidden: 64                          # Number of hidden units for encoder model f (?)
f_output_features: 64                 # Number of output features for encoder model f (?)
f_dropout: 0.1                        # Dropout for encoder model f (?)
f_layers: 2                           # Number of layers for encoder model f (?)
k_hidden: 64                          # Number of hidden units for adversary model k (?)
k_output_features: 64                 # Number of output features for adversary model k (?)
k_dropout: 0.1                        # Dropout in GCN for adversary model k
k_nlayer: 2                           # Number of layers in GCN for adversary model k
c_hidden: 64                          # Number of hidden units for classifier model (?)
c_input: 64                           # Number of input units for classifier model (?)
alpha: 10                             # Loss alpha weight ...
beta: 0.1                             # Loss beta weight ...
gamma: 0.1                            # Loss gamma weight ...
lam: 0.5                              # Loss lambda weight ...
k_lr: 0.001                           # Learning rate for adversary model k
c_lr: 0.001                           # Learining rate for classifier
g_lr: 0.0001                          # Learning rate for joint f and g models optimization
g_warmup_lr: 0.001                    # Learning rate for a warmup optimizer for augmentation model g
f_lr: 0.001                           # Learning rate for encoder model f
graphair_temperature: 0.07            # Temperature for the whole graphair model
edge_perturbation: True               # Whether to use edge perturbation
node_feature_masking: True            # Whether to use feature masking