-
Notifications
You must be signed in to change notification settings - Fork 1
/
tune.py
133 lines (113 loc) · 6.41 KB
/
tune.py
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
import argparse
import torch
from torch import optim
from torchdrug import models, tasks, core
from torchdrug.layers import distribution
from utils import load_dataset, load_GSN_dataset, load_GNN
def finetune_GCPN(dataset, gnn_type, task, pretrained_model_path, num_epoch=10):
model = load_GNN(dataset=dataset, model_type='GCPN', gnn_type=gnn_type)
if task == 'plogp':
finetune_task = tasks.GCPNGeneration(model, dataset.atom_types,
max_edge_unroll=12, max_node=38,
task='plogp', criterion='ppo',
reward_temperature=1,
agent_update_interval=3,
gamma=0.9)
elif task == 'qed':
finetune_task = tasks.GCPNGeneration(model, dataset.atom_types,
max_edge_unroll=12, max_node=38,
task='qed', criterion=('ppo', 'nll'),
reward_temperature=1,
agent_update_interval=3,
gamma=0.9)
else:
raise ValueError(f'Unknown task {task}')
finetune_optimizer = optim.Adam(finetune_task.parameters(), lr=1e-5)
finetune_solver = core.Engine(finetune_task, dataset, None, None,
finetune_optimizer, gpus=(0,),
batch_size=64, log_interval=200)
finetune_solver.load(pretrained_model_path, load_optimizer=False)
finetune_solver.train(num_epoch=num_epoch)
finetune_model_path = pretrained_model_path.replace('.pickle', f'_finetune_{task}.pickle')
finetune_solver.save(finetune_model_path)
finetune_results = finetune_task.generate(num_sample=32)
print(finetune_results.to_smiles())
if task == 'plogp':
print(finetune_task.best_results['Penalized logP'])
elif task == 'qed':
print(finetune_task.best_results['QED'])
else:
raise ValueError(f'Unknown task {task}')
def finetune_GraphAF(dataset, gnn_type, task, pretrained_model_path, num_epoch=5):
model = load_GNN(dataset=dataset, model_type='GraphAF', gnn_type=gnn_type)
num_atom_type = dataset.num_atom_type
num_bond_type = dataset.num_bond_type + 1 # add one class for non-edge
node_prior = distribution.IndependentGaussian(torch.zeros(num_atom_type),
torch.ones(num_atom_type))
edge_prior = distribution.IndependentGaussian(torch.zeros(num_bond_type),
torch.ones(num_bond_type))
node_flow = models.GraphAF(model, node_prior, num_layer=12)
edge_flow = models.GraphAF(model, edge_prior, use_edge=True, num_layer=12)
if task == 'plogp':
finetune_task = tasks.AutoregressiveGeneration(node_flow, edge_flow,
max_edge_unroll=12, max_node=38,
task='plogp', criterion='ppo',
reward_temperature=20,
baseline_momentum=0.9,
agent_update_interval=5,
gamma=0.9)
elif task == 'qed':
finetune_task = tasks.AutoregressiveGeneration(node_flow, edge_flow,
max_edge_unroll=12, max_node=38,
task='qed', criterion={'ppo': 0.25, 'nll': 1.0},
reward_temperature=10,
baseline_momentum=0.9,
agent_update_interval=5,
gamma=0.9)
else:
raise ValueError(f'Unknown task {task}')
finetune_optimizer = optim.Adam(finetune_task.parameters(), lr=1e-5)
finetune_solver = core.Engine(finetune_task, dataset, None, None,
finetune_optimizer, gpus=(0,),
batch_size=64, log_interval=200)
finetune_solver.load(pretrained_model_path, load_optimizer=False)
finetune_solver.train(num_epoch=num_epoch)
finetune_model_path = pretrained_model_path.replace('.pickle', f'_finetune_{task}.pickle')
finetune_solver.save(finetune_model_path)
finetune_results = finetune_task.generate(num_sample=32)
print(finetune_results.to_smiles())
if task == 'plogp':
print(finetune_task.best_results['Penalized logP'])
elif task == 'qed':
print(finetune_task.best_results['QED'])
else:
raise ValueError(f'Unknown task {task}')
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--model_type', type=str, required=True)
parser.add_argument('--gnn_type', type=str, required=True)
parser.add_argument('--task', type=str, required=True)
parser.add_argument('--data_dir', type=str, required=True)
parser.add_argument('--pretrained_model_path', type=str, required=True)
parser.add_argument('--num_epoch', type=int, default=-1)
args = parser.parse_args()
if args.gnn_type == 'GSN':
dataset = load_GSN_dataset(args.data_dir)
else:
dataset = load_dataset(args.data_dir)
if args.num_epoch == -1:
if args.model_type == 'GCPN':
finetune_GCPN(dataset=dataset, gnn_type=args.gnn_type, task=args.task,
pretrained_model_path=args.pretrained_model_path)
elif args.model_type == 'GraphAF':
finetune_GraphAF(dataset=dataset, gnn_type=args.gnn_type, task=args.task,
pretrained_model_path=args.pretrained_model_path)
else:
if args.model_type == 'GCPN':
finetune_GCPN(dataset=dataset, gnn_type=args.gnn_type, task=args.task,
pretrained_model_path=args.pretrained_model_path,
num_epoch=args.num_epoch)
elif args.model_type == 'GraphAF':
finetune_GraphAF(dataset=dataset, gnn_type=args.gnn_type, task=args.task,
pretrained_model_path=args.pretrained_model_path,
num_epoch=args.num_epoch)