-
Notifications
You must be signed in to change notification settings - Fork 14
/
Copy pathauswahlen.py
190 lines (174 loc) · 7.24 KB
/
auswahlen.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
190
# -*- coding: utf-8 -*-
"""
Created on Tue Aug 4 10:12:03 2020
@author: wu
"""
# COPYRIGHT:
# AUTH: jian.wu
# Email:fengyuguohou2010@hotmail.com
#!pip install varclushi
from varclushi import VarClusHi
def psi_cut(A,DA,x):
A.replace(['non','none','None','NONE','null','NULL','Null','"null"','[]','[ ]','{}','{ }',' ','nu',np.nan,float('inf')],0,inplace=True)
A=A[[x]]
bins=[-float('inf')]+DA
bins=list(set(bins))
bins.sort()
a=pd.DataFrame([bins[-1]]).T
a.columns =A.columns
A=pd.concat([A,a],ignore_index=True)
if float('inf') in bins:
bins.remove(float('inf'))
lab=[i for i in bins[1:]]
AA=pd.DataFrame()
AA[x] = pd.cut(A[x], bins, labels = lab)
AA[x] = AA[x].astype('float64')
df=pd.DataFrame()
df=AA[x].groupby(AA[x]).agg(['count'])
df[x+'_count']=df['count']
return df[[x+'_count']],AA[[x]]
def psi(A,B,col=[],name='',p='',ll=True,lll=False,l1=400,l2=700):
if p!='':
if len(A)<len(B):
BB=B.sample(len(A))
AA=A
else:
AA=A.sample(len(B))
BB=B
p_bin=[l1]+[l1+i*5 for i in range(120)]
A_1,A_3=psi_cut(A,p_bin,p)
B_1,B_3=psi_cut(B,p_bin,p)
A_4,A_2=psi_cut(AA,p_bin,p)
B_4,B_2=psi_cut(BB,p_bin,p)
AA_1=A_1.copy()
BB_1=B_1.copy()
AA_1['pp']=np.cumsum(A_1/A_1.sum())
BB_1['pp']=np.cumsum(B_1/B_1.sum())
plt.figure(figsize = (6, 3))
plt.title('score'+'_psi')
if lll:
AA_1['pp'].plot(linewidth=1.5,xlim=(l1,l2),label='A',color='red')
BB_1['pp'].plot(linewidth=1.5,xlim=(l1,l2),label='B',color='blue')
else:
A_2[p].plot(kind='density',linewidth=1.5,xlim=(l1,l2),label='A',color='red')
B_2[p].plot(kind='density',linewidth=1.5,xlim=(l1,l2),label='B',color='blue')
plt.legend(loc='upper right')
plt.show()
BX=B_1-A_1
BX=BX.fillna(0)
BX=BX-BX
A_1=A_1-BX
A_1=A_1.fillna(0)
B_1=B_1-BX
B_1=B_1.fillna(0)
A_1=A_1+1
B_1=B_1+1
ss=(-A_1/A_1.sum()+B_1/B_1.sum())*np.log((B_1/B_1.sum())/(A_1/A_1.sum()))
ss.replace(['non','none','None','NONE','null','NULL','Null','"null"','[]','[ ]','{}','{ }',' ','nu',np.nan,float('inf')],0,inplace=True)
s=float(ss.sum())
mmx=list(A_1[A_1[p+'_count']==float(np.max(A_1))].index).pop()
A_1_1=A_1[A_1.index>=mmx]
A_1_2=A_1[A_1.index<mmx]
B_1_1=B_1[B_1.index>=mmx]
B_1_2=B_1[B_1.index<mmx]
ps1=(A_1_1/A_1_1.sum()-B_1_1/B_1_1.sum())
ps1.replace(['non','none','None','NONE','null','NULL','Null','"null"','[]','[ ]','{}','{ }',' ','nu',np.nan,float('inf')],0,inplace=True)
pss1=float(ps1.sum())
ps2=(B_1_2/B_1_2.sum()-A_1_2/A_1_2.sum())
ps2.replace(['non','none','None','NONE','null','NULL','Null','"null"','[]','[ ]','{}','{ }',' ','nu',np.nan,float('inf')],0,inplace=True)
pss2=float(ps2.sum())
return s,pss1+pss2
elif name!='':
col=[float(min(A[name]))]+[float(min(A[name]))+(float(max(A[name]))-float(min(A[name])))*i/20 for i in range(21)]
col1=[float(min(B[name]))]+[float(min(B[name]))+(float(max(B[name]))-float(min(B[name])))*i/20 for i in range(21)]
col=col+col1+[max(B[name])]+[max(A[name])]
a=(float(max(A[name]))-float(min(A[name])))/30
b=(float(max(B[name]))-float(min(B[name])))/30
x=min(a,b)
col.sort()
ccc=[]
ccc.append(col[-1])
for i in range(len(col)-1):
if col[-i]-col[-i-1]>x:
ccc.append(col[-i-1])
A_1,A_2=psi_cut(A,ccc,name)
B_1,B_2=psi_cut(B,ccc,name)
BX=B_1-A_1
BX=BX.fillna(0)
BX=BX-BX
A_1=A_1-BX
A_1=A_1.fillna(0)
B_1=B_1-BX
B_1=B_1.fillna(0)
A_1=A_1+1
B_1=B_1+1
A_1_1=A_1/np.float(A_1.sum())
B_1_1=B_1/np.float(B_1.sum())
ccc=list(A_1_1.index)+list(B_1_1.index)
ccc=list(set(ccc))
c=pd.DataFrame(index=ccc)
c[0]=A_1_1+0.00001
c[1]=B_1_1+0.00001
c=c.fillna(0.00001)
if ll:
plt.figure(figsize = (8, 4))
plt.subplot(1, 1, 1)
plt.title(name+'_psi')
bar_width = 0.35
plt.bar(np.arange(len(A_1_1)),list(A_1_1[name+'_count']),bar_width,label='A',color='red')
plt.bar(np.arange(len(B_1_1))+bar_width, list(B_1_1[name+'_count']),bar_width ,label='B')
if len(A_1_1)>=len(B_1_1):
plt.xticks(np.arange(len(A_1_1))+bar_width/2,tuple(A_1_1.index))
else:
plt.xticks(np.arange(len(B_1_1))+bar_width/2,tuple(B_1_1.index))
plt.legend(loc='upper center', fancybox=True, ncol=5)
plt.show()
ss=(c[0]-c[1])*np.log(c[0]/c[1])
ss.replace(['non','none','None','NONE','null','NULL','Null','"null"','[]','[ ]','{}','{ }',' ','nu',np.nan,float('inf')],0,inplace=True)
s=float(ss.sum())
return s
def ksss(df, score_name, y_name):
df=df.sort_values(by=score_name)
grouped = df[y_name].groupby(df[score_name])
df_agg = grouped.agg(['count', 'sum'])
df_agg = df_agg.sort_index(ascending = False)
df_agg['good'] = df_agg['count'] - df_agg['sum']
df_agg['cum_bad_rate'] = np.cumsum(df_agg['sum']) / sum(df_agg['sum'])
df_agg['cum_good_rate'] = np.cumsum(df_agg['good']) / sum(df_agg['good'])
return np.max(abs(df_agg['cum_bad_rate'] - df_agg['cum_good_rate']))
class auswahle():
def __init__(self,data,test_data,y):
self.data = data
self.test_data=test_data
self.y = y
def psi_ceshi(self,col):
cc=[]
for l in col:
cc.append(psi(self.data ,self.test_data,col=[],name=l,p='',ll=False,lll=False,l1=400,l2=700))
ppsi=pd.DataFrame(index=range(len(col)))
ppsi['var']=col
ppsi['psi']=cc
return ppsi
def del_high_corr_ks(self,model,col,p_cri,y):
corr = self.data[col].corr(method = 'pearson')
corr_vars = corr.sum()
corr_vars=corr_vars.sort_values()
var = list(corr_vars.index)
var_del = []
for i in range(len(var) - 1):
var_i = var[i]
if var_i not in var_del:
for j in range(i+1,len(var)):
if j!=1:
var_j = var[j]
if var_j not in var_del:
if (abs(corr[var_i][var_j]) > p_cri):
if (ksss(self.data,var_i,y) >= ksss(self.data,var_j,y)):
var_del.append(var_j)
else:
var_del.append(var_i)
col1=list(set(var) - set(var_del))
return col1
def varclust(self,col):
varclust = VarClusHi(self.data[col])
return varclust,demo1_vc.info,demo1_vc.rsquare