-
Notifications
You must be signed in to change notification settings - Fork 3
/
cs_lassologit_predict.do
253 lines (167 loc) · 5.54 KB
/
cs_lassologit_predict.do
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
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
clear all
********************************************************************************
*** Verify post-logit predictions ***
********************************************************************************
$loadspam
gen myholdout = _n>4000
foreach lam in 0.18 .1 .05 .01 0.005 0.001 {
cap drop yhat* phat*
lassologit v58 v1-v57, l(`lam') postlogit lambdan
predict double yhat_my, xb
predict double phat_my, pr
local sel = e(selected)
logit v58 `sel'
predict double yhat_logit, xb
predict double phat_logit, pr
assert reldif(yhat_my,yhat_logit)<10e4
assert reldif(phat_my,phat_logit)<10e4
}
//
********************************************************************************
*** Verify "predict" using internal ("Mata") predictions ***
********************************************************************************
foreach lam in 0.18 .1 .05 .01 0.005 0.001 {
$loadspam
gen myholdout = _n>4000
di "lambda = `lam'"
lassologit v58 v1-v57, l(`lam') savep lambdan
mat P = e(phat)
svmat P
predict double p0, pr
assert reldif(P1,p0)<10e-4
}
//
foreach lam in 0.18 .1 .05 .01 0.005 0.001 {
$loadspam
gen myholdout = _n>4000
di "lambda = `lam'"
lassologit v58 v1-v57, l(`lam') savep postlogit lambdan
mat P = e(phat)
svmat P
predict double p0, pr postlogit
assert reldif(P1,p0)<10e-4
}
//
foreach lam in .1 .05 .01 0.005 0.001 {
$loadspam
gen myholdout = _n>4000
di "lambda = `lam'"
lassologit v58 v1-v57, l(`lam') savep nocons lambdan
mat P = e(phat)
svmat P
predict double p0, pr
assert reldif(P1,p0)<10e-4
}
//
********************************************************************************
*** Verify using glmnet (cs_predict.R) ***
********************************************************************************
$loadprostate
gen ybin = lpsa > 2
lassologit ybin lcavol-pgg45, lam(.2 .11 .1 0.05 0.01) savep lambdan
mat P = e(phat)
mat P = P[97,1..5]
mat G = ( 0.761871934040682,0.883722877193935,0.895110432984738,0.955542318252603,0.999505561515738)
assert mreldif(P,G)<0.01
lassologit ybin lcavol-pgg45, lam(.2 .11 .1 0.05 0.01) savep nocons lambdan
mat P = e(phat)
mat P = P[97,1..5]
mat G = ( 0.761411339295924,0.883320216213757,0.893762132556822,0.956213536648691,0.999298395412146)
assert mreldif(P,G)<0.01
lassologit ybin lcavol-pgg45, lam(.2 .11 .1 0.05 0.01) savep nostd lambdan
mat P = e(phat)
mat P = P[97,1..5]
mat G = ( 0.708638828605104,0.858851204259658,0.872566014019627,0.931492511316991,0.998846180327787)
assert mreldif(P,G)<0.01
lassologit ybin lcavol-pgg45, lam(.2 .11 .1 0.05 0.01) savep nostd nocons lambdan
mat P = e(phat)
mat P = P[97,1..5]
mat G = ( 0.703648308982215,0.853973607193486,0.868497657957821,0.929886289492582,0.998267904285371)
assert mreldif(P,G)<0.01
********************************************************************************
*** comparison with mata predicted values ***
********************************************************************************
*** in-sample
$loadspam
gen myholdout = _n<50
lassologit v58 v1-v57, lambda(0.1) savep lambdan
mat Phat1 = e(phat)
svmat Phat1
predict double Phat2 if e(sample), pr
assert abs(Phat1 - Phat2)<10e-8
*** in-sample (not on full sample)
$loadspam
gen myholdout = _n<50
lassologit v58 v1-v57 if _n<4000, lambda(0.1) savep lambdan
mat Phat1 = e(phat)
svmat Phat1
predict double Phat2 if e(sample), pr
assert abs(Phat1 - Phat2)<10e-6 if !missing(Phat1)
*** in-sample
$loadspam
gen myholdout = _n<50
lassologit v58 v1-v57, lambda(0.1) savep holdout(myh) lambdan
mat Phat1 = e(phat0)
svmat Phat1
predict double Phat2, pr
assert abs(Phat1 - Phat2)<10e-8 if _n<50
********************************************************************************
*** predicted values: re-estimation and no re-estimation ***
********************************************************************************
$loadprostate
gen ybin = lpsa > 2
lassologit ybin lcavol-pgg45
predict double yhat , lic(ebic)
predict double yhat2, lic(ebic) est
assert abs(yhat- yhat2)<10e-6
cap drop yhat yhat2
********************************************************************************
*** more testing ***
********************************************************************************
$loadprostate
gen ybin = lpsa > 2
* post logit
cap drop yhat yhat2
lassologit ybin lcavol-pgg45
predict double yhat , lic(ebic) postlogit // postlogit is being ignored
predict double yhat2, lic(ebic) est postlogit
** cvlassologit: compare two methods
cap drop yhat*
set seed 1
cvlassologit ybin lcavol-pgg45
cvlassologit, lopt postres
predict double yhat1
set seed 1
cvlassologit ybin lcavol-pgg45
predict double yhat2, lopt
assert abs(yhat1- yhat2)<10e-6
** cvlassologit with postlogit: compare two methods
cap drop yhat*
set seed 1
cvlassologit ybin lcavol-pgg45
cvlassologit, lopt postres postlogit
predict double yhat1, postlogit
set seed 1
cvlassologit ybin lcavol-pgg45
predict double yhat2, lopt postlogit
assert abs(yhat1- yhat2)<10e-6
** rlassologit
cap drop yhat*
rlassologit ybin lcavol-pgg45
predict double yhat, postlogit
** lassologit: compare two methods
cap drop yhat*
lassologit ybin lcavol-pgg45
lassologit, lic(ebic) postres
predict double yhat1
lassologit ybin lcavol-pgg45
predict double yhat2, lic(ebic)
assert abs(yhat1- yhat2)<10e-6
** lassologit with post-lasso: compare two methods
cap drop yhat*
lassologit ybin lcavol-pgg45
lassologit, lic(ebic) postres postlogit
predict double yhat1, postlogit
lassologit ybin lcavol-pgg45
predict double yhat2, lic(ebic) postlogit est
assert abs(yhat1- yhat2)<10e-6