PatchNet/main_single.py at main · zjunet/PatchNet · 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
import numpy as np
import torch
import torch.optim as optim
from config import args
from dataloader import DataLoaderSubstructContext, DataLoaderMasking
from util import ExtractSubstructureContextPair, cycle_index, MaskAtom

from datasets import MoleculeDataset
from tqdm import tqdm
import time
from multimodel import PretrainModule
import min_norm_solvers
from ogb.graphproppred import PygGraphPropPredDataset
from ogb.lsc import PygPCQM4Mv2Dataset
from ogb.utils import smiles2graph

def train(args, device, loader_CP, loader_AM, optimizer):
    curr_losses = {}
    model.train()
    model.zero_grad()
    Loss=0
    step=0
    for batch_CP, batch_AM in zip(tqdm(loader_CP), loader_AM):
        step += 1
        optimizer.zero_grad()
        loss_data = {}
        grads = {}
        batch_CP.to(device)
        batch_AM.to(device)
        batch = {'CP': batch_CP, 'AM': batch_AM}
        tasks = {'CP', 'AM'}
        # -------------- Begin of Pareto Multi-Tasking Learning --------------
        if 'CP' in tasks:
            t = 'CP'
            loss = model.CP(batch[t])
            grads[t] = []
            loss_data[t] = loss.data
            loss.backward()
            for param in model.main_model.parameters():
                if param.grad is not None:
                    grads[t].append(param.grad.data.detach().cpu())
            model.zero_grad()

        if 'AM' in tasks:
            t = 'AM'
            loss = model.AM(batch[t])
            grads[t] = []
            loss_data[t] = loss.data
            loss.backward()
            for param in model.main_model.parameters():
                if param.grad is not None:
                    grads[t].append(param.grad.data.detach().cpu())
            model.zero_grad()

        if len(tasks) > 1:
            gn = min_norm_solvers.gradient_normalizers(grads, loss_data, 'l2')
            for t in loss_data:
                for gr_i in range(len(grads[t])):
                    grads[t][gr_i] = grads[t][gr_i] / gn[t].to(grads[t][gr_i].device)
            sol, _ = min_norm_solvers.MinNormSolver.find_min_norm_element_FW([grads[t] for t in tasks])
            sol = {k: sol[i] for i, k in enumerate(tasks)}
        # -------------- End of Pareto Multi-Tasking Learning --------------
        model.zero_grad()
        train_loss = 0
        actual_loss = 0
        loss_dict = model(batch)

        for i, l in loss_dict.items():
            train_loss += float(sol[i]) * l
            actual_loss += l

        train_loss.backward()

        loss_dict['train_loss'] = actual_loss.detach()
        for k, v in sol.items():
            loss_dict[k + '_weight'] = torch.tensor(float(v))
            if k not in curr_losses:
                curr_losses[k] = loss_dict[k].item()
            else:
                curr_losses[k] += loss_dict[k].item()
        if 'train_loss' not in curr_losses:
            curr_losses['train_loss'] = loss_dict['train_loss']
        else:
            curr_losses['train_loss'] += loss_dict['train_loss']

        optimizer.step()
        model.zero_grad()

        Loss += curr_losses['train_loss']
        for k in curr_losses:
            curr_losses[k] = 0


    return Loss / step

if __name__ == '__main__':

    np.random.seed(0)
    torch.manual_seed(0)
    device = torch.device('cuda:' + str(args.device)) \
        if torch.cuda.is_available() else torch.device('cpu')
    if torch.cuda.is_available():
        torch.cuda.manual_seed_all(0)
        torch.cuda.set_device(args.device)

    l1 = args.num_layer - 1
    l2 = l1 + args.csize
    print('num layer: %d l1: %d l2: %d' % (args.num_layer, l1, l2))

    if args.pretrain_dataset=='zinc':
        dataset_CP = MoleculeDataset(args.dataset_root + args.dataset, dataset=args.dataset,
                                  transform=ExtractSubstructureContextPair(args.num_layer, l1, l2))
        dataset_AM = MoleculeDataset(args.dataset_root + args.dataset, dataset=args.dataset,
                              transform=MaskAtom(num_atom_type=119, num_edge_type=5,
                                                 mask_rate=args.mask_rate, mask_edge=args.mask_edge))
    elif args.pretrain_dataset=='pcba':
        dataset_CP = PygGraphPropPredDataset(name='ogbg-molpcba', root='/data/syf/finetune/dataset/',
                                         transform=ExtractSubstructureContextPair(args.num_layer, l1, l2))
        dataset_AM = PygGraphPropPredDataset(name='ogbg-molpcba', root='/data/syf/finetune/dataset/',
                            transform=MaskAtom(num_atom_type=119, num_edge_type=5,
                                               mask_rate=args.mask_rate, mask_edge=args.mask_edge))
    elif args.pretrain_dataset=='pcqm':
        dataset_CP = PygPCQM4Mv2Dataset(root='/data/syf/finetune/dataset/', smiles2graph=smiles2graph,
                                    transform=ExtractSubstructureContextPair(args.num_layer, l1, l2))
        dataset_AM = PygPCQM4Mv2Dataset(root='/data/syf/finetune/dataset/', smiles2graph=smiles2graph,
                                    transform=MaskAtom(num_atom_type=119, num_edge_type=5,
                                                       mask_rate=args.mask_rate, mask_edge=args.mask_edge))
    if args.pretrain_dataset=='':
        raise 'Pretrain Dataset Invalid'

    loader_CP = DataLoaderSubstructContext(dataset_CP, batch_size=args.batch_size,
                                           shuffle=False, num_workers=args.num_workers)
    loader_AM = DataLoaderMasking(dataset_AM, batch_size=args.batch_size,
                                  shuffle=False, num_workers=args.num_workers)

    model = PretrainModule(args.gnn_type, args.num_layer, args.win_size, args.step,
                        args.emb_dim * 2, args.emb_dim, args.for_dropout,
                        args.dropout, args.gnn_dropout, args.num_heads, args.gat_heads,
                        args.pooling, l1, l2, args.token_size, args.num_tokens, args.pretrain_dataset,
                        args.k, args.sim_function, args.sparse, args.activation_learner, args.thresh).to(device)

    model_param_group = [{'params': model.parameters(), 'lr': args.lr}]
    optimizer = optim.Adam(model_param_group, lr=args.lr, weight_decay=args.decay)

    for epoch in range(1, args.epochs + 1):
        print('epoch: {}'.format(epoch))
        train_loss = train(args, device, loader_CP, loader_AM, optimizer)
        print(f'epoch{epoch}: loss={train_loss}')
        if not args.output_model_dir == '':
            torch.save(model.main_model.state_dict(),
                       args.output_model_dir + 'Multie' + str(epoch) + '_model.pth')

    if not args.output_model_dir == '':
        torch.save(model.main_model.state_dict(),
                   args.output_model_dir + 'Multi_model.pth')