Catalog
One 、 Code organization and execution sequence 3
Two 、 Problem description 3
3、 ... and 、 feature extraction 4
（ One ）dat_risk 4
（ Two ）dat_symbol 4
（ 3、 ... and ）dat_app 6
（ Four ）dat_edge 7
Four 、 Feature extraction of user association graph 13
（ One ） Centrality class features 13
（ Two ）Louvain Community clustering 13
5、 ... and 、 Tag spread 14
（ One ） Once a contact 14
（ Two ） Second degree contact 15
（ 3、 ... and ） Once routed 17
6、 ... and 、 Characteristic propagation 17
（ One ） Feature Overview 18
（ Two ） Data description 18
（ 3、 ... and ） feature extraction 18
（ Four ） Characteristic evaluation 18
7、 ... and 、 Data summary and characteristic comparison 18
8、 ... and 、 Feature sorting and discretization 19
Nine 、 Adjustable parameter 19
Ten 、 Test set AUC The mystery of decline 21
（ One ） user app The difference in the missing rate of data 21
（ Two ） The degree to which the user's associated data is closely related 21
11、 ... and 、 Improved space 22
One 、 Code organization and execution sequence
Two 、 Problem description
The explanation here is to reach a consensus , At the same time, it is also to unify the symbolic representation , The following explanations are based on this . The data about user characteristics is divided into four parts （ Sort according to the difficulty of handling , From easy to difficult ）:
•（1）dat_risk
•（2）dat_symbol
•（3）dat_app
•（4）dat_edge
About user tags and training sets 、 Verification set 、 Test set data : （1）sample_train( Two : id、label) （2）valid_id( A column of : id) （3）test_id( A column of : id)
hold sample_train、valid_id、test_id Of id Splice up , Get everything id, The data name is recorded as all_id, The data format is a 28959*1 Of DataFrame.
3、 ... and 、 feature extraction
（ One ）dat_risk
Output is all_id_dat_risk
hold dat_risk and all_id Make internal connections :
all_id_dat_risk = pd.merge(all_id, dat_risk, on=‘id’, how=‘inner’)
The aim is to find out all_id Every one of them id Characteristics of , Of course there are id Probably not , In the final data, it is shown as missing value .
Reprinted from ：http://www.biyezuopin.vip/onews.asp?id=16293
