def main():
dataset, learning_rate, epoch, alpha, beta, gamma, pretrain_epoch = parse_argument(
)
random.seed(9001)
dataset_config = {
'feature_file': './Database/' + dataset + '/features.txt',
'graph_file': './Database/' + dataset + '/edges.txt',
'walks_file': './Database/' + dataset + '/walks.txt',
'label_file': './Database/' + dataset + '/group.txt'
}
graph = Dataset(dataset_config)
pretrain_config = {
'net_shape': [6, 1],
'att_shape': [1],
'net_input_dim': graph.num_nodes,
'att_input_dim': graph.num_feas,
'pretrain_params_path': './Log/' + dataset + '/pretrain_params.pkl'
}
model_config = {
'net_shape': [6, 1],
'att_shape': [1],
'net_input_dim': graph.num_nodes,
'att_input_dim': graph.num_feas,
'is_init': True,
'pretrain_params_path': './Log/' + dataset + '/pretrain_params.pkl'
}
trainer_config = {
'net_shape': [6, 1],
'att_shape': [1],
'net_input_dim': graph.num_nodes,
'att_input_dim': graph.num_feas,
'drop_prob': 0.2,
'learning_rate': learning_rate,
'batch_size': 100,
'num_epochs': epoch,
'beta': beta,
'alpha': alpha,
'gamma': gamma,
'model_path': './Log/' + dataset + '/test_model.pkl',
}
print("=" * 30 + "开始进行参数预训练" + "=" * 30)
pretrainer = PreTrainer(pretrain_config)
pretrainer.pretrain(graph.X, 'net', pretrain_epoch) # walks.txt的矩阵
pretrainer.pretrain(graph.Z, 'att', pretrain_epoch)
model = Model(model_config)
trainer = Trainer(model, trainer_config)
trainer.train(graph)
train_emb = trainer.infer(graph)
train_emb = np.array(train_emb)
with open("./Log/" + dataset + "/train_emb.txt", 'w') as f:
np.savetxt(f, train_emb)
ccsistent_loss,
'beta_W':
beta_W,
'View_num':
View_num,
'View':
layers,
'dims':
dims,
'drop_prob':
0.2,
'learning_rate':
learning_rate,
'batch_size':
2000,
'num_epochs':
1000,
'model_path':
'./Log/' + dataset_name + '/' + dataset_name + '_model.pkl',
}
model = MVModel(model_config)
trainer = Trainer(model, trainer_config)
trainer.train(graph)
acc, nmi = trainer.inferCluster(graph)
result_single = 'ccsistent_loss={:.4f}:'.format(
ccsistent_loss) + ' acc={:.4f}'.format(
acc) + ' & ' + 'nmi={:.4f}'.format(nmi)
f.write(result_single + '\n')
f.flush()
test_mode=True)
val_loader = torch.utils.data.DataLoader(val_dataset,
batch_size=64,
shuffle=False,
num_workers=0)
print('num_train_files: ' + str(len(train_dataset.filepaths)))
print('num_val_files: ' + str(len(val_dataset.filepaths)))
trainer = Trainer(cnet,
train_loader,
val_loader,
optimizer,
nn.CrossEntropyLoss(),
'svcnn',
log_dir,
num_views=1)
trainer.train(30)
# STAGE 2
log_dir = args.name + '_stage_2'
create_folder(log_dir)
cnet_2 = MVCNN(args.name,
cnet,
nclasses=40,
cnn_name=args.cnn_name,
num_views=args.num_views)
del cnet
optimizer = optim.Adam(cnet_2.parameters(),
lr=args.lr,
weight_decay=args.weight_decay,
betas=(0.9, 0.999))
