-
Notifications
You must be signed in to change notification settings - Fork 19
/
test.py
189 lines (168 loc) · 6.76 KB
/
test.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
185
186
187
188
189
'''
论文引用关系预测
'''
import pandas as pd
import numpy as np
import sys,os,pickle,nltk
from sklearn.cross_validation import StratifiedShuffleSplit
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.decomposition import TruncatedSVD
from scipy.spatial import distance
from scipy.stats import pearsonr
import math
import networkx as nx
import lightgbm as lgb
from collections import Counter
from sklearn.metrics import log_loss,accuracy_score,f1_score
sys.path.append(os.path.abspath('../src'))
'''配置项
'''
folder_name='data/raw/'
train_file='data/raw/training_set.txt'
test_file='data/raw/testing_set.txt'
output_file='data/raw/output.txt'
'''全局变量
'''
train_df=pd.read_csv(train_file,sep=' ',names=['sid','tid','label'])
test_df=pd.read_csv(test_file,sep=' ',names=['sid','tid','label'])
node_df=pd.read_csv('data/raw/node_information.csv',names=['id','year','title','authors','journal','abstract'])
features={}
'''工具函数
'''
def cosine(a,b):
res=0
if np.linalg.norm(a)!=0 and np.linalg.norm(b)!=0:
res=distance.cosine(a,b)
return res
def get_vectors(a_list,b_list):
'''将两个list转换为两个one-hot向量
'''
id2index=dict([(id,i) for i,id in enumerate(set(a_list+b_list))])
a=np.zeros((len(id2index),))
b=np.zeros((len(id2index),))
for key in a_list:
a[id2index[key]]=1
for key in b_list:
b[id2index[key]]=1
return a,b
def prepare_node_df():
'''对node_df进行预置处理
'''
nltk.download('punkt') # for tokenization
nltk.download('stopwords')
stpwds = set(nltk.corpus.stopwords.words("english"))
stemmer = nltk.stem.PorterStemmer()
#分割作者字段
node_df.authors.fillna('',inplace=True)
node_df.loc[:,'author_set']=node_df.authors.apply(lambda x:[x.strip() for x in x.lower().split(',') if len(x.strip())>0])
#分割文章标题
node_df.title.fillna('',inplace=True)
node_df.loc[:,'title_set']=node_df.title.apply(lambda x:set([stemmer.stem(w) for w in x.lower().split(' ') if w not in stpwds]))
#处理摘要
node_df.abstract.fillna('',inplace=True)
node_df.loc[:,'simple_abstract']=node_df.abstract.apply(lambda x:' '.join([w.strip() for w in x.lower().split(' ') if w not in stpwds]))
#处理journal
node_df.journal.fillna('',inplace=True)
node_df.loc[:,'journal_set']=node_df.journal.apply(lambda x:set([w.strip() for w in x.split('.') if len(x.strip())>0]))
def append_node_index(df):
'''向df中添加source index及target index
依赖全局变量：node_df
'''
id2index=dict([(id,i) for i,id in enumerate(node_df.id)])
df.loc[:,'sindex']=df.sid.apply(lambda x:id2index[x])
df.loc[:,'tindex']=df.tid.apply(lambda x:id2index[x])
def set_feature(name,val):
'''设置特征值
'''
if name not in features:
features[name]=[]
features[name].append(val)
def get_features(df):
'''特征工程
'''
features.clear()
for i,row in enumerate(df.itertuples()):
source=node_df.loc[row.sindex]
target=node_df.loc[row.tindex]
# 年份特征
set_feature('year_dis',source.year-target.year)
set_feature('year_source',source.year)
set_feature('year_target',target.year)
# 共同作者数量
set_feature('common_author',len(set(source.author_set).intersection(set(target.author_set))))
# 标题中共现词数
set_feature('overlap_title',len(source.title_set.intersection(target.title_set)))
# 期刊中共现词数
set_feature('overlap_journal',len(source.journal_set.intersection(target.journal_set)))
#摘要的cosine相似度
set_feature('abstract_tfidf_similar',cosine(abstract_tfidf[row.sindex].toarray()[0],abstract_tfidf[row.tindex].toarray()[0]))
set_feature('abstract_svd_similar',cosine(abstract_svd[row.sindex],abstract_svd[row.tindex]))
# 节点的度
set_feature('in_degree_target',in_degree.get(row.tid,0))
set_feature('out_degree_source',out_degree.get(row.sid,0))
# jaccard index
sn=G.neighbors(row.sid)
tn=G.neighbors(row.tid)
common_ns=len(set(sn).intersection(set(tn)))
set_feature('g_jaccard_index',common_ns/(1e-6+len(set(sn+tn))))
sn_vec,tn_vec=get_vectors(sn,tn)
set_feature('g_neighbour_sqrt',common_ns/(math.sqrt(len(sn)+len(tn))+1e-6))
set_feature('g_neighbour_cosine',distance.cosine(sn_vec,tn_vec))
set_feature('g_neighbour_pearson',pearsonr(sn_vec,tn_vec)[0])
set_feature('g_cluster_source',g_cluster.get(row.sid,0))
set_feature('g_cluster_target',g_cluster.get(row.tid,0))
set_feature('g_kcore_source',g_kcore.get(row.sid,0))
set_feature('g_kcore_target',g_kcore.get(row.tid,0))
set_feature('g_pagerank_source',g_pagerank.get(row.sid,0))
set_feature('g_pagerank_target',g_pagerank.get(row.tid,0))
set_feature('g_aver_neighbour_source',g_aver_neighbor.get(row.sid,0))
set_feature('g_aver_neighbour_target',g_aver_neighbor.get(row.tid,0))
if i%10000==0:
print(i,"training examples processsed")
feature_df=pd.DataFrame(data=features)
return feature_df
if __name__=='__main__':
print('正在抽取特征...')
prepare_node_df()
append_node_index(train_df)
append_node_index(test_df)
'''文本tfidf特征
'''
vectorizer=TfidfVectorizer(min_df=2)
abstract_tfidf=vectorizer.fit_transform(node_df.simple_abstract)
abstract_svd=TruncatedSVD(n_components=100,random_state=100).fit_transform(abstract_tfidf)
'''构建有向图
'''
DiG=nx.DiGraph()
G=nx.Graph()
for row in train_df.itertuples():
DiG.add_edge(row.sid,row.tid)
G.add_edge(row.sid,row.tid)
in_degree=DiG.in_degree()
out_degree=DiG.out_degree()
'''graph dict
'''
g_cluster=nx.algorithms.cluster.clustering(G)
g_kcore=nx.core_number(G)
g_pagerank=nx.pagerank(G)
g_aver_neighbor=nx.average_neighbor_degree(G)
train_xs=get_features(train_df)
test_xs=get_features(test_df)
'''lightgbm
'''
gbm = lgb.LGBMClassifier(objective='binary',
num_leaves=31,
learning_rate=0.05,
n_estimators=1000,subsample=0.8,)
gbm.fit(train_xs.values, train_df.label,verbose=200)
test_ys=gbm.predict(test_xs.values)
train_xs.to_csv(folder_name+'/train_xs.csv',index=False)
test_xs.to_csv(folder_name+'/test_xs.csv',index=False)
'''输出结果
'''
with open(output_file,'w') as f:
f.write('Id,prediction\n')
for i,val in enumerate(test_ys):
f.write('%d,%d\n'%(i,val))
print('预测结果输出到',output_file)
print('结果输出占比',Counter(test_ys))