GFM-BA/preprocess.py at main · RingBDStack/GFM-BA · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
import torch
from utils.data_utils import *
from model.gcope import get_model
from torch_geometric.datasets import TUDataset, Planetoid, Amazon, Coauthor, Reddit

def gcope_data_preprocess(args, dataset):
    with torch.no_grad():
        data, gco_model, raw_data = get_clustered_data(dataset, args.data_path_origin, args.cross_link, args.unify_dim, args.cl_init_method)
    return data, gco_model, raw_data


def mdgpt_data_preprocess(args, dataset, pretrain=True, combine=True, sample=True):
    if pretrain:
        print('Loading pretrain data')
        feature_list = []
        adj_list = []
        for data_name in dataset:
            if data_name in ['Cora', 'CiteSeer', 'PubMed']:
                data = Planetoid(root=args.data_path_origin, name=data_name)._data
            elif data_name in ['Computers', 'Photo']:
                data = Amazon(root=args.data_path_origin, name=data_name)._data
            else:
                raise ValueError(f'Unknown dataset: {data_name}')

            feature, adj = process_tu(data)
            feature = pca_compression(feature, k=args.unify_dim)
            feature = torch.FloatTensor(feature).to(args.device)

            adj = normalize_adj(adj + sp.eye(adj.shape[0]))
            if args.sparse:
                adj = sparse_mx_to_torch_sparse_tensor(adj)
            adj = adj.to(args.device)

            feature_list.append(feature)
            adj_list.append(adj)

        if combine or sample:
            print('Combining, it takes long...')
            adj = combine_dataset(*adj_list)
        if sample:
            print('Negative sampling, it takes long...')
            negetive_sample = prompt_pretrain_sample(adj, 50).to(args.device)
            return feature_list, adj_list, negetive_sample

        return feature_list, adj_list

    else:
        if dataset in ['Cora', 'CiteSeer', 'PubMed']:
            data = Planetoid(root=args.data_path_origin, name=dataset)._data
        elif dataset in ['Computers', 'Photo']:
            data = Amazon(root=args.data_path_origin, name=dataset)._data
        else:
            raise ValueError(f'Unknown dataset: {dataset}')

        feature, adj = process_tu(data)
        feature = pca_compression(feature, k=args.unify_dim)
        feature = torch.FloatTensor(feature).to(args.device)

        adj = normalize_adj(adj + sp.eye(adj.shape[0]))
        if args.sparse:
            adj = sparse_mx_to_torch_sparse_tensor(adj)
        adj = adj.to(args.device)

        if not hasattr(data, 'train_mask'):
            print('Manual split train/val/test')


            train_indices = []
            val_indices = []
            test_indices = []

            test_indices = range(int(data.y.shape[0] - args.test_idx_num), data.y.shape[0])

            remaining_indices = list(set(range(data.y.shape[0])) - set(test_indices))
            random.shuffle(remaining_indices)

            val_size = int(len(remaining_indices) * 0.1)
            train_size = len(remaining_indices) - val_size

            train_indices = remaining_indices[:train_size]
            val_indices = remaining_indices[train_size:]

            num_nodes = data.y.size(0)
            data.train_mask = torch.zeros(num_nodes, dtype=torch.bool)
            data.val_mask = torch.zeros(num_nodes, dtype=torch.bool)
            data.test_mask = torch.zeros(num_nodes, dtype=torch.bool)

            for idx in train_indices:
                data.train_mask[idx] = True
            for idx in val_indices:
                data.val_mask[idx] = True
            for idx in test_indices:
                data.test_mask[idx] = True


        labels = data.y.to(args.device)
        idx_train = torch.nonzero(data.train_mask).squeeze()
        idx_test = torch.nonzero(data.test_mask).squeeze()
        return feature, adj, labels, idx_train, idx_test


def samgpt_data_preprocess(args, pretrain_dataset_names, version=0):
    def load_dataset(name, path=args.data_path_origin):
        if name in ['Cora', 'CiteSeer', 'PubMed']:
            dataset = Planetoid(root=path, name=name)
        elif name in ['Computers', 'Photo']:
            dataset = Amazon(root=path, name=name)
        else:
            raise ValueError(f"Unknown dataset name: {name}")
        return dataset

    if version == 0:
        pretrain_loaders = [DataLoader(load_dataset(dataset)) for dataset in pretrain_dataset_names]
        aug_features = []
        aug_adjs = []
        lbls = []
        for step, datas in enumerate(zip(*pretrain_loaders)):
            for pretrain_dataset_name, data in zip(pretrain_dataset_names, datas):
                feature, adj = process_tu(data)
                feature = torch.FloatTensor(pca_compression(feature, k=args.unify_dim))

                if not(os.path.exists(f'./data/samgpt/{pretrain_dataset_name}_aug_feature.pt') and \
                        os.path.exists(f'./data/samgpt/{pretrain_dataset_name}_aug_adj.pt') and \
                        os.path.exists(f'./data/samgpt/{pretrain_dataset_name}_lbl.pt') ):
                    aug_feature, aug_adj, lbl = build_aug(adj, feature, args.sparse, args.drop_percent)
                    torch.save(aug_feature, f'./data/samgpt/{pretrain_dataset_name}_aug_feature.pt')
                    torch.save(aug_adj, f'./data/samgpt/{pretrain_dataset_name}_aug_adj.pt')
                    torch.save(lbl,  f'./data/samgpt/{pretrain_dataset_name}_lbl.pt')

                aug_feature, aug_adj, lbl = torch.load(f'./data/samgpt/{pretrain_dataset_name}_aug_feature.pt'), \
                torch.load(f'./data/samgpt/{pretrain_dataset_name}_aug_adj.pt'),  \
                torch.load(f'./data/samgpt/{pretrain_dataset_name}_lbl.pt')
                aug_features.append(aug_feature)
                aug_adjs.append(aug_adj)
                lbls.append(lbl)

        aug_features = [tensors.to(args.device) for tensors in aug_features]
        aug_adjs = [tensors.to(args.device) for tensors in aug_adjs]
        lbls = [tensors.to(args.device) for tensors in lbls]

        return aug_features, aug_adjs, lbls

    elif version == 1:
        pretrain_loaders = [DataLoader(load_dataset(dataset)) for dataset in pretrain_dataset_names]
        feature_list = []
        adj_list = []
        for step, datas in enumerate(zip(*pretrain_loaders)):
            for pretrain_dataset_name, data in zip(pretrain_dataset_names, datas):
                feature, adj = process_tu(data)
                feature = torch.FloatTensor(pca_compression(feature, k=args.unify_dim))
                adj = normalize_adj(adj + sp.eye(adj.shape[0]))
                feature_list.append(feature)
                adj_list.append(adj)

        feature_list = [tensors.to(args.device) for tensors in feature_list]
        adj_list = [tensors.to(args.device) for tensors in adj_list]

        return feature_list, adj_list

    elif version == 2:
        pretrain_datasets = [load_dataset(dataset) for dataset in pretrain_dataset_names]
        feature_list, edge_index_list = [], []
        for data in pretrain_datasets:
            data = data._data
            feature_list.append(data.x)
            edge_index_list.append(data.edge_index)
        return feature_list, edge_index_list


from model.mdgpt import PrePrompt
from model.samgpt import samgpt_PrePrompt
def get_pretrain_model(args):
    if args.gfm_model == "GCOPE":
        model = get_model(num_features=args.unify_dim, hid_dim=args.hid_dim, num_conv_layers=args.num_conv_layers, dropout=args.dropout)
    elif args.gfm_model == "MDGPT":
        model = PrePrompt(args.unify_dim, args.hid_units, args.nonlinearity, 3, 0.1, args.combinetype).to(args.device)
    elif args.gfm_model == "SAMGPT":
        model = samgpt_PrePrompt(args.unify_dim,
                                 args.hid_units,
                                 args.nonlinearity,
                                 args.num_pretrain_dataset_num,
                                 args.num_layers_num,
                                 0.1,
                                 type_=args.combinetype,
                                 backbone=args.backbone,
                                 alpha=args.alpha,
                                 ablation=args.ablation_pre).to(args.device)
    else:
        raise NotImplementedError
    return model