forked from lyjps/Struct2GO
-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_struct2GO2.py
184 lines (163 loc) · 7.94 KB
/
train_struct2GO2.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
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
from cProfile import label
from random import shuffle
from re import T
from statistics import mode
import numpy as np
import torch
from torch.utils.data import Dataset, DataLoader
import pickle
from model.network import SAGNetworkHierarchical2,SAGNetworkGlobal
import torch.nn as nn
import torch.optim as optim
import dgl
from dgl.dataloading import GraphDataLoader
import torch.nn.functional as F
from tkinter import _flatten
from sklearn import metrics
from sklearn.metrics import roc_auc_score, roc_curve, auc, precision_score, recall_score, f1_score, average_precision_score
import argparse
import warnings
import datetime
import matplotlib.pyplot as plt
from sklearn.metrics import precision_recall_curve
from data_processing.divide_data import MyDataSet
from model.evaluation import cacul_aupr,calculate_performance
from transformers import get_cosine_schedule_with_warmup
from tqdm import tqdm
import logging
import os
def create_logger(branch_name):
logger = logging.getLogger(branch_name)
handler1 = logging.StreamHandler()
handler2 = logging.FileHandler(filename=os.path.join('log',branch_name+'2.log'))
logger.setLevel(logging.DEBUG)
handler1.setLevel(logging.ERROR)
handler2.setLevel(logging.DEBUG)
formatter = logging.Formatter("%(asctime)s %(name)s %(levelname)s %(message)s")
handler1.setFormatter(formatter)
handler2.setFormatter(formatter)
logger.addHandler(handler1)
logger.addHandler(handler2)
return logger
warnings.filterwarnings('ignore')
Thresholds = [x/100 for x in range(1,100)]
if __name__ == "__main__":
#参数设置
device = "cuda:7"
parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument('-batch_size', '--batch_size', type=int, default=64, help="the number of the bach size")
parser.add_argument('-learningrate', '--learningrate',type=float,default=5e-5)
parser.add_argument('-dropout', '--dropout',type=float,default=0.5)
parser.add_argument('-branch', '--branch',type=str,default='mf')
parser.add_argument('-labels_num', '--labels_num',type=int,default=328)
args = parser.parse_args()
# 根据选择的标签大类自动填写训练文件路径
train_data_path = 'divided_data/'+args.branch+'_train_dataset'
valid_data_path = 'divided_data/'+args.branch+'_valid_dataset'
label_network_path = 'processed_data/label_'+args.branch+'_network'
logger = create_logger(args.branch)
with open(train_data_path,'rb')as f:
train_dataset = pickle.load(f)
with open(valid_data_path,'rb')as f:
valid_dataset = pickle.load(f)
with open(label_network_path,'rb')as f:
label_network=pickle.load(f)
label_network = label_network.to(device)
label_topo_order_list = dgl.topological_nodes_generator(label_network)
# 载入/设置参数
epoch_num = 20
batch_size = args.batch_size
learningrate = args.learningrate
dropout = args.dropout
labels_num = args.labels_num
# 加载数据
train_dataloader = GraphDataLoader(dataset=train_dataset, batch_size = batch_size, drop_last = False, shuffle = True)
valid_dataloader = GraphDataLoader(dataset=valid_dataset, batch_size = batch_size, drop_last = False, shuffle = True)
# TODO: 这里的输入特征应该是one-hot(26)+node2vec(30) = 56
# 但暂时没做特征拼接，先用node2vec特征试试
model = SAGNetworkHierarchical2(30, 512, labels_num, num_convs=6, pool_ratio=0.75, dropout=dropout).to(device)
optimizer = optim.Adam(model.parameters(), lr=learningrate)
lr_scheduler = get_cosine_schedule_with_warmup(optimizer, num_warmup_steps=100, num_training_steps=epoch_num*len(train_dataloader))
criterion = nn.CrossEntropyLoss()
best_fscore = 0
best_aupr = 0
best_scores = []
best_score_dict = {}
logger.info('#########'+args.branch+'###########')
logger.info('########start training###########')
for epoch in range(epoch_num):
print("epoch:",epoch)
logger.info("epoch: "+str(epoch))
model.train()
train_loss = 0
print("training")
logger.info("training")
for i,(graphs, labels, seq_feats) in tqdm(enumerate(train_dataloader)):
graphs = graphs.to(device)
seq_feats = seq_feats.to(device)
labels = labels.to(device)
labels = torch.squeeze(labels)
if len(labels.shape)==1:
labels = labels.unsqueeze(0)
optimizer.zero_grad()
logits = model(graphs,seq_feats,label_network,label_topo_order_list)
loss = criterion(logits,labels)
loss.backward()
optimizer.step()
lr_scheduler.step()
# 累加计算平均loss
train_loss+=loss.item()
if i%30 == 29:
logger.info(f'Epoch: {epoch} / {epoch_num}, Step: {i} / {len(train_dataloader)}, Loss(batch): {loss.item()}')
# 每四轮进行一次验证
if epoch%4==3:
model.eval()
print("validating")
logger.info("validating")
valid_loss = 0
pred = []
actual = []
with torch.no_grad():
for i,(graphs, labels, seq_feats) in tqdm(enumerate(valid_dataloader)):
graphs = graphs.to(device)
seq_feats = seq_feats.to(device)
labels = labels.to(device)
labels = torch.squeeze(labels)
if len(labels.shape)==1:
labels = labels.unsqueeze(0)
logits = model(graphs,seq_feats,label_network,label_topo_order_list)
logits = F.sigmoid(logits)
loss = criterion(logits,labels)
valid_loss += loss.item()
pred += logits.tolist()
actual += labels.tolist()
if i%10 == 9:
logger.info(f'Valid Step: {i} / {len(valid_dataloader)}, Loss(batch): {loss.item()}')
fpr, tpr, th = roc_curve(np.array(actual).flatten(), np.array(pred).flatten(), pos_label=1)
auc_score = auc(fpr, tpr)
aupr = cacul_aupr(np.array(actual).flatten(), np.array(pred).flatten())
score_dict = {}
each_best_fcore = 0
each_best_scores = []
for thresh in tqdm(Thresholds):
f_score, precision, recall = calculate_performance(actual, pred, label_network,threshold=thresh)
if f_score >= each_best_fcore:
each_best_fcore = f_score
each_best_scores = [thresh, f_score, recall, precision, auc_score]
scores = [f_score, recall, precision, auc_score]
score_dict[thresh] = scores
if each_best_fcore >= best_fscore:
best_fscore = each_best_fcore
best_scores = each_best_scores
best_score_dict = score_dict
best_aupr = aupr
torch.save(model, 'save_models/bestmodel2_{}_{}_{}_{}.pkl'.format(args.branch,batch_size,learningrate,dropout))
thresh, f_score, recall = each_best_scores[0], each_best_scores[1], each_best_scores[2]
precision, auc_score = each_best_scores[3], each_best_scores[4]
logger.info('########valid metric###########')
logger.info('epoch{}, train_loss{}, valid_loss:{}'.format(epoch, train_loss/len(train_dataloader), valid_loss/len(valid_dataloader)))
logger.info('threshold:{}, f_score{}, auc{}, recall{}, precision{}, aupr{}'.format(thresh, f_score, auc_score, recall, precision, aupr))
logger.info('best_fscore: '+str(best_fscore))
logger.info('best_scores[thresh,fmax,recall,precision,auc]: '+str(best_scores))
logger.info('best_aupr: '+str(best_aupr))
logger.info('best_score_dict: '+str(best_score_dict))