-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.py
163 lines (137 loc) · 6.17 KB
/
main.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
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
import torch
import os
import ray
import time
import dill
import random
import argparse
import datetime
import numpy as np
from utils import *
from train import *
from dataset import *
from models.baseline import DotProd
from models.mlp import MLP
from models.e2e import GraphAutoEncoder_e2e
from torch_geometric.data import DataLoader
from model_selection import model_selection, split
# Set random seed
seed = 9
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
#ray.init() # local ray initialization
ray.init(address=os.environ.get("ip_head"), _redis_password=os.environ.get("redis_password")) # clustering ray initialization
print('Settings:')
print('\tKMP_SETTING:', os.environ.get('KMP_SETTING'))
print('\tOMP_NUM_THREADS:', os.environ.get('OMP_NUM_THREADS'))
print('\tKMP_BLOCKTIME:', os.environ.get('KMP_BLOCKTIME'))
print('\tMALLOC_CONF:', os.environ.get('MALLOC_CONF'))
print('\tLD_PRELOAD:', os.environ.get('LD_PRELOAD'))
if __name__ == "__main__":
t0 = time.time()
parser = argparse.ArgumentParser()
parser.add_argument('--dataset-path', dest='dataset_path')
parser.add_argument('--name', dest='name')
parser.add_argument('--mode', dest='mode', default='Validation', choices=['Validation', 'Test'])
parser.add_argument('--k', dest='k', default=5, type=float)
parser.add_argument('--epochs', dest='epochs', default=10000, type=int)
parser.add_argument('--batch', dest='batch', default=512, type=int)
parser.add_argument('--csv', dest='csv_path')
parser.add_argument('--e2e', dest='e2e', action='store_true')
parser.add_argument('--baseline', dest='baseline', required=False, choices=['MLP', 'DotProd'])
parser.add_argument('--tag', dest='grouped_stratification', action='store_true')
args = parser.parse_args()
assert args.baseline or args.e2e, "Params e2e and baseline cannot be set simultaniously"
data = DrugRepurposing(args.dataset_path)
# Split data into Traingin and Test
if args.grouped_stratification:
train_ids, test_ids = split(y=data['y'], eval_size=0.2, groups=data['tag'], seed=seed)
groups = data.getitem_from_key(train_ids, 'tag')
else:
train_ids, test_ids = split(y=data['y'], eval_size=0.2, seed=seed)
groups = None
dill.dump((test_ids, data.getitem_from_key(test_ids, 'drug')), open('test_drug_name_'+args.name+'.p', 'wb'))
x, _ = data[0]
if not args.baseline:
model = MLP() if not args.e2e else GraphAutoEncoder_e2e(protein_size=x.size(0))
else:
model = MLP() if args.baseline == 'MLP' else DotProd()
train_dataloader = DataLoader(data[train_ids], batch_size=args.batch, shuffle=True)
test_dataloader = DataLoader(data[test_ids], batch_size=args.batch)
device = torch.device("cpu")
if args.mode == 'Validation' and args.k is not None:
# Run model selection
if not args.k:
args.k = 0.2
if args.e2e:
# Structural-Similarity-based prediction Network (SSN)
configurations = {
'mlp_output_size': [1],
'mlp_hidden_size': [[512, 128, 32], [512, 64],
[256, 64, 16], [256,32],
[128, 64, 32], [128,16]],
'mlp_batchnorm': [False, True],
'ae_batchnorm': [False],
'denoising': [True],
'gnn_num_layers': [3],
'gnn_hidden_size':[192],
'gnn_output_size':[192],
'lr': [2e-5, 2e-4, 2e-3]
}
elif args.baseline == 'DotProd':
prot_dim = data.data[0]['gene_emb'].size(0)
drug_dim = data.data[0]['drug_emb'].size(0)
# Baseline DotProd(MLP(node2vec emb), MLP(MorganFP))
configurations = {
'prot_dim': [prot_dim],
'drug_dim': [drug_dim],
'hidden_dim': [4096, 3000, 2048, 1024, 512, 256, 128, 64, 32],
'lr': [2e-5, 2e-4, 2e-3]
}
else:
# Chemical-Similarity-based prediction Network (CSN)
# and baseline concat(node2vec emb, MorganFP) + MLP
configurations = {
'input_size': [x.size(0)], 'output_size': [1],
'hidden_size': [[512, 128, 32], [512, 64], [512],
[256, 64, 16], [256,32], [256],
[128, 64, 32], [128,16], [128]],
'use_batchnorm': [True, False],
'lr': [2e-5, 2e-4, 2e-3]
}
max_tasks = 100
conf = model_selection(args.k, configurations, model, data[train_ids], args.epochs, groups, args.batch, args.csv_path, device=device, max_concurrent_tasks=max_tasks)
else:
conf = {
'mlp_output_size': [1],
'mlp_hidden_size': [512, 64],
'mlp_batchnorm': [False],
'ae_batchnorm': [False],
'denoising': [True],
'gnn_num_layers': [3],
'gnn_hidden_size':[192],
'gnn_output_size':[192],
'lr': 2e-3
}
'''
conf = {
'input_size': x.size(0), 'output_size': 1,
'hidden_size': [512, 128, 32],
'lr': 2e-5
}
'''
# Train the best model on the whole training set and evaluate on the test set
remote_id = train_and_eval.remote(model, conf, args.epochs, train_dataloader, test_dataloader, device,
mode='Test', save_best=True, path_save_best="best_"+args.name+".pth")
history = ray.get(remote_id)
dill.dump({"epochs": args.epochs,
"batch_size": args.batch,
'model_structure': conf,
'model_selection': configurations if args.mode == 'Validation' else None,
"history": history}, open(args.name + "_history.p", "wb"))
title = args.name.replace("_", " ")
make_plots(history['history'], title, args.name, 'Test')
elapsed = time.time() - t0
print(str(datetime.timedelta(seconds=int(round((elapsed))))))