in python/dgllife/model/pretrain/moleculenet/pcba.py [0:0]
def create_pcba_model(model_name):
"""Create a model.
Parameters
----------
model_name : str
Name for the model.
Returns
-------
Created model
"""
n_tasks = 128
if model_name == 'GCN_canonical_PCBA':
num_gnn_layers = 2
dropout = 0.053320999462421345
return GCNPredictor(in_feats=74,
hidden_feats=[128] * num_gnn_layers,
activation=[F.relu] * num_gnn_layers,
residual=[True] * num_gnn_layers,
batchnorm=[True] * num_gnn_layers,
dropout=[dropout] * num_gnn_layers,
predictor_hidden_feats=1024,
predictor_dropout=dropout,
n_tasks=n_tasks)
elif model_name == 'GCN_attentivefp_PCBA':
num_gnn_layers = 5
dropout = 0.24826461148671453
return GCNPredictor(in_feats=39,
hidden_feats=[128] * num_gnn_layers,
activation=[F.relu] * num_gnn_layers,
residual=[True] * num_gnn_layers,
batchnorm=[True] * num_gnn_layers,
dropout=[dropout] * num_gnn_layers,
predictor_hidden_feats=64,
predictor_dropout=dropout,
n_tasks=n_tasks)
elif model_name == 'GAT_canonical_PCBA':
num_gnn_layers = 1
dropout = 0.008451521225305653
return GATPredictor(in_feats=74,
hidden_feats=[64] * num_gnn_layers,
num_heads=[8] * num_gnn_layers,
feat_drops=[dropout] * num_gnn_layers,
attn_drops=[dropout] * num_gnn_layers,
alphas=[0.0194367227727808] * num_gnn_layers,
residuals=[False] * num_gnn_layers,
biases=[False] * num_gnn_layers,
predictor_hidden_feats=64,
predictor_dropout=dropout,
n_tasks=n_tasks)
elif model_name == 'GAT_attentivefp_PCBA':
num_gnn_layers = 1
dropout = 0.2811043226878611
return GATPredictor(in_feats=39,
hidden_feats=[32] * num_gnn_layers,
num_heads=[8] * num_gnn_layers,
feat_drops=[dropout] * num_gnn_layers,
attn_drops=[dropout] * num_gnn_layers,
alphas=[0.25837424873685433] * num_gnn_layers,
residuals=[True] * num_gnn_layers,
biases=[False] * num_gnn_layers,
predictor_hidden_feats=16,
predictor_dropout=dropout,
n_tasks=n_tasks)
elif model_name == 'Weave_canonical_PCBA':
return WeavePredictor(node_in_feats=74,
edge_in_feats=13,
num_gnn_layers=1,
gnn_hidden_feats=128,
graph_feats=256,
gaussian_expand=True,
n_tasks=n_tasks)
elif model_name == 'Weave_attentivefp_PCBA':
return WeavePredictor(node_in_feats=39,
edge_in_feats=11,
num_gnn_layers=2,
gnn_hidden_feats=64,
graph_feats=64,
gaussian_expand=False,
n_tasks=n_tasks)
elif model_name == 'MPNN_canonical_PCBA':
return MPNNPredictor(node_in_feats=74,
edge_in_feats=13,
node_out_feats=32,
edge_hidden_feats=64,
num_step_message_passing=1,
num_step_set2set=3,
num_layer_set2set=3,
n_tasks=n_tasks)
elif model_name == 'MPNN_attentivefp_PCBA':
return MPNNPredictor(node_in_feats=39,
edge_in_feats=11,
node_out_feats=64,
edge_hidden_feats=32,
num_step_message_passing=2,
num_step_set2set=1,
num_layer_set2set=1,
n_tasks=n_tasks)
elif model_name == 'AttentiveFP_canonical_PCBA':
return AttentiveFPPredictor(node_feat_size=74,
edge_feat_size=13,
num_layers=2,
num_timesteps=3,
graph_feat_size=32,
dropout=0.05370268638522968,
n_tasks=n_tasks)
elif model_name == 'AttentiveFP_attentivefp_PCBA':
return AttentiveFPPredictor(node_feat_size=39,
edge_feat_size=11,
num_layers=3,
num_timesteps=2,
graph_feat_size=16,
dropout=0.31957324617702254,
n_tasks=n_tasks)
elif model_name == 'gin_supervised_contextpred_PCBA':
jk = 'last'
model = GINPredictor(
num_node_emb_list=[120, 3],
num_edge_emb_list=[6, 3],
num_layers=5,
emb_dim=300,
JK=jk,
dropout=0.5,
readout='attention',
n_tasks=n_tasks
)
model.gnn.JK = jk
return model
elif model_name == 'gin_supervised_infomax_PCBA':
jk = 'concat'
model = GINPredictor(
num_node_emb_list=[120, 3],
num_edge_emb_list=[6, 3],
num_layers=5,
emb_dim=300,
JK=jk,
dropout=0.5,
readout='max',
n_tasks=n_tasks
)
model.gnn.JK = jk
return model
elif model_name == 'gin_supervised_edgepred_PCBA':
jk = 'sum'
model = GINPredictor(
num_node_emb_list=[120, 3],
num_edge_emb_list=[6, 3],
num_layers=5,
emb_dim=300,
JK=jk,
dropout=0.5,
readout='attention',
n_tasks=n_tasks
)
model.gnn.JK = jk
return model
elif model_name == 'gin_supervised_masking_PCBA':
jk = 'sum'
model = GINPredictor(
num_node_emb_list=[120, 3],
num_edge_emb_list=[6, 3],
num_layers=5,
emb_dim=300,
JK=jk,
dropout=0.5,
readout='attention',
n_tasks=n_tasks
)
model.gnn.JK = jk
return model
else:
return None