FAMILY-package: A Convex Formulation for Modeling Interactions with Strong...
In FAMILY: A Convex Formulation for Modeling Interactions with Strong Heredity
Description
Details
Author(s)
References
See Also
Examples
Description
Fits linear and logistic regression models with pairwise interactions for two sets of covariates. The models fitted by this package obey strong heredity which implies that interaction of two covariates is only included if both main effects are included in the model. In the case of high dimensional data we may not wish to model all possible covariates and their interactions but a small subset of them, thus we utilize the penalized regression methods of Haris, Witten and Simon (2014).
Details
Package: FAMILY
Type: Package
Version: 0.1.19
Date: 2015-06-20
License: GPL-2
The package includes the following functions:
FAMILY: Fit a penalized regression model for a grid of alpha and lambda values
predict.FAMILY: Predict yhat or phat for a for a given data set from the fitted model
coef.FAMILY: Extract the set of estimated non-zero coefficients
Author(s)
Asad Haris, Daniela Witten and Noah Simon
Maintainer: Asad Haris <aharis@uw.edu>
References
Haris, Witten and Simon (2014). Convex Modeling of Interactions with Strong Heredity. Available on ArXiv at http://arxiv.org/abs/1410.3517
See Also
FAMILY,
predict.FAMILY,
coef.FAMILY
Examples
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
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
library(FAMILY)
library(pROC)
library(pheatmap)
#####################################################################################
#####################################################################################
############################# EXAMPLE - CONTINUOUS RESPONSE #########################
#####################################################################################
#####################################################################################
############################## GENERATE DATA ########################################
#Generate training set of covariates X and Z
set.seed(1)
X.tr<- matrix(rnorm(10*100),ncol = 10, nrow = 100)
Z.tr<- matrix(rnorm(15*100),ncol = 15, nrow = 100)
#Generate test set of covariates X and Z
X.te<- matrix(rnorm(10*100),ncol = 10, nrow = 100)
Z.te<- matrix(rnorm(15*100),ncol = 15, nrow = 100)
#Scale appropiately
meanX<- apply(X.tr,2,mean)
meanY<- apply(Z.tr,2,mean)
X.tr<- scale(X.tr, scale = FALSE)
Z.tr<- scale(Z.tr, scale = FALSE)
X.te<- scale(X.te,center = meanX,scale = FALSE)
Z.te<- scale(Z.te,center = meanY,scale = FALSE)
#Generate full matrix of Covariates
w.tr<- c()
w.te<- c()
X1<- cbind(1,X.tr)
Z1<- cbind(1,Z.tr)
X2<- cbind(1,X.te)
Z2<- cbind(1,Z.te)
for(i in 1:16){
for(j in 1:11){
w.tr<- cbind(w.tr,X1[,j]*Z1[,i])
w.te<- cbind(w.te, X2[,j]*Z2[,i])
}
}
#Generate response variables with signal from
#First 5 X features and 5 Z features.
#We construct the coefficient matrix B.
#B[1,1] contains the intercept
#B[-1,1] contains the main effects for X.
# For instance, B[2,1] is the main effect for the first feature in X.
#B[1,-1] contains the main effects for Z.
# For instance, B[1,10] is the coefficient for the 10th feature in Z.
#B[i+1,j+1] is the coefficient of X_i Z_j
B<- matrix(0,ncol = 16,nrow = 11)
rownames(B)<- c("inter" , paste("X",1:(nrow(B)-1),sep = ""))
colnames(B)<- c("inter" , paste("Z",1:(ncol(B)-1),sep = ""))
# First, we simulate data as follows:
# The first five features in X, and the first five features in Z, are non-zero.
# And given the non-zero main effects, all possible interactions are involved.
# We call this "high strong heredity"
B_high_SH<- B
B_high_SH[1:6,1:6]<- 1
#View true coefficient matrix
pheatmap(as.matrix(B_high_SH), scale="none",
cluster_rows=FALSE, cluster_cols=FALSE)
Y_high_SH <- as.vector(w.tr%*%as.vector(B_high_SH))+rnorm(100,sd = 2)
Y_high_SH.te <- as.vector(w.te%*%as.vector(B_high_SH))+rnorm(100,sd = 2)
# Now a new setting:
# Again, the first five features in X, and the first five features in Z, are involved.
# But this time, only a subset of the possible interactions are involved.
# Strong heredity is still maintained.
# We call this "low strong heredity"
B_low_SH<- B_high_SH
B_low_SH[2:6,2:6]<-0
B_low_SH[3:4,3:5]<- 1
#View true coefficient matrix
pheatmap(as.matrix(B_low_SH), scale="none",
cluster_rows=FALSE, cluster_cols=FALSE)
Y_low_SH <- as.vector(w.tr%*%as.vector(B_low_SH))+rnorm(100,sd = 1.5)
Y_low_SH.te <- as.vector(w.te%*%as.vector(B_low_SH))+rnorm(100,sd = 1.5)
############################## FIT SOME MODELS ########################################
#Define alphas and lambdas
#Define 3 different alpha values
#Low alpha values penalize groups more
#High alpha values penalize individual Interactions more
alphas<- c(0.01,0.5,0.99)
lambdas<- seq(0.1,1,length = 50)
#high Strong heredity with l2 norm
fit_high_SH<- FAMILY(X.tr, Z.tr, Y_high_SH, lambdas ,
alphas, quad = TRUE,iter=500, verbose = TRUE )
yhat_hSH<- predict(fit_high_SH, X.te, Z.te)
mse_hSH <-apply(yhat_hSH,c(2,3), "-" ,Y_high_SH.te)
mse_hSH<- apply(mse_hSH^2,c(2,3),sum)
#Find optimal model and plot matrix
im<- which(mse_hSH==min(mse_hSH),TRUE)
plot(fit_high_SH$Estimate[[im[2] ]][[im[1]]])
#Plot some matrices for different alpha values
#Low alpha, higher penalty on groups
plot(fit_high_SH$Estimate[[ 1 ]][[ 25 ]])
#Medium alpha, equal penalty on groups and individual interactions
plot(fit_high_SH$Estimate[[ 2 ]][[ 25 ]])
#High alpha, more penalty on individual interactions
plot(fit_high_SH$Estimate[[ 3 ]][[ 40 ]])
#View Coefficients
coef(fit_high_SH)[[im[2]]][[im[1]]]
############################## Uncomment code for EXAMPLE ###########################
# #high Strong heredity with l_infinity norm norm
# fit_high_SH<- FAMILY(X.tr, Z.tr, Y_high_SH, lambdas ,
# alphas, quad = TRUE,iter=500, verbose = TRUE,
# norm = "l_inf")
# yhat_hSH<- predict(fit_high_SH, X.te, Z.te)
# mse_hSH <-apply(yhat_hSH,c(2,3), "-" ,Y_high_SH.te)
# mse_hSH<- apply(mse_hSH^2,c(2,3),sum)
#
# #Find optimal model and plot matrix
# im<- which(mse_hSH==min(mse_hSH),TRUE)
# plot(fit_high_SH$Estimate[[im[2] ]][[im[1]]])
#
#
# #Plot some matrices for different alpha values
# #Low alpha, higher penalty on groups
# plot(fit_high_SH$Estimate[[ 1 ]][[ 30 ]])
# #Medium alpha, equal penalty on groups and individual interactions
# plot(fit_high_SH$Estimate[[ 2 ]][[ 10 ]])
# #High alpha, more penalty on individual interactions
# plot(fit_high_SH$Estimate[[ 3 ]][[ 20 ]])
#
#
# #View Coefficients
# coef(fit_high_SH)[[im[2]]][[im[1]]]
############################## Uncomment code for EXAMPLE ###########################
# #Redefine lambdas
# lambdas<- seq(0.1,0.5,length = 50)
#
# #low Strong heredity with l_2 norm
# fit_low_SH<- FAMILY(X.tr, Z.tr, Y_low_SH, lambdas ,
# alphas, quad = TRUE,iter=500, verbose = TRUE )
# yhat_lSH<- predict(fit_low_SH, X.te, Z.te)
# mse_lSH <-apply(yhat_lSH,c(2,3), "-" ,Y_low_SH.te)
# mse_lSH<- apply(mse_lSH^2,c(2,3),sum)
#
# #Find optimal model and plot matrix
# im<- which(mse_lSH==min(mse_lSH),TRUE)
# plot(fit_low_SH$Estimate[[im[2] ]][[im[1]]])
#
#
# #Plot some matrices for different alpha values
# #Low alpha, higher penalty on groups
# plot(fit_low_SH$Estimate[[ 1 ]][[ 25 ]])
# #Medium alpha, equal penalty on groups and individual interactions
# plot(fit_low_SH$Estimate[[ 2 ]][[ 10 ]])
# #High alpha, more penalty on individual interactions
# plot(fit_low_SH$Estimate[[ 3 ]][[ 10 ]])
#
#
# #View Coefficients
# coef(fit_low_SH)[[im[2]]][[im[1]]]
#####################################################################################
#####################################################################################
############################### EXAMPLE - BINARY RESPONSE ###########################
#####################################################################################
#####################################################################################
############################## GENERATE DATA ########################################
#Generate data for logistic regression
Yp_high_SH<- as.vector((w.tr)%*%as.vector(B_high_SH))
Yp_high_SH.te<- as.vector((w.te)%*%as.vector(B_high_SH))
Yprobs_high_SH<- 1/(1+exp(-Yp_high_SH))
Yprobs_high_SH.te<- 1/(1+exp(-Yp_high_SH.te))
Yp_high_SH<- rbinom(100, size = 1, prob = Yprobs_high_SH)
Yp_high_SH.te<- rbinom(100, size = 1, prob = Yprobs_high_SH.te)
lambdas<- seq(0.01,0.15,length = 50)
############################## FIT SOME MODELS ########################################
#Fit glm via l_2 norm
fit_high_SH<- FAMILY(X.tr, Z.tr, Yp_high_SH, lambdas ,
alphas, quad = TRUE,iter=500, verbose = TRUE,
family = "binomial")
yhp_hSH<- predict(fit_high_SH, X.te, Z.te)
mse_high_SH <-apply(yhp_hSH,c(2,3), "-" ,Yp_high_SH.te)
mse_hSH<- apply(mse_high_SH^2,c(2,3),sum)
im<- which(mse_hSH==min(mse_hSH),TRUE)
plot(fit_high_SH$Estimate[[im[2] ]][[im[1]]])
roc( Yp_high_SH.te,yhp_hSH[,im[1],im[2]],plot = TRUE)
#View Coefficients
coef(fit_high_SH)[[im[2]]][[im[1]]]
############################## Uncomment code for EXAMPLE ###########################
# #Fit glm via l_infinity norm
# fit_high_SH<- FAMILY(X.tr, Z.tr, Yp_high_SH, lambdas , norm = "l_inf",
# alphas, quad = TRUE,iter=500, verbose = TRUE,
# family = "binomial")
# yhp_hSH<- predict(fit_high_SH, X.te, Z.te)
# mse_high_SH <-apply(yhp_hSH,c(2,3), "-" ,Yp_high_SH.te)
# mse_hSH<- apply(mse_high_SH^2,c(2,3),sum)
# im<- which(mse_hSH==min(mse_hSH),TRUE)
# plot(fit_high_SH$Estimate[[im[2] ]][[im[1]]])
# roc( Yp_high_SH.te,yhp_hSH[,im[1],im[2]],plot = TRUE)
#
# #View Coefficients
# coef(fit_high_SH)[[im[2]]][[im[1]]]
#####################################################################################
#####################################################################################
############################## EXAMPLE WHERE X=Z ####################################
######################## Uncomment Code for EXAMPLE #################################
#####################################################################################
############################## GENERATE DATA ########################################
# #Redefine Lambdas
# lambdas<- seq(0.01,0.3,length = 50)
#
#
# #We consider the case X=Z now
# w.tr<- c()
# w.te<- c()
# X1<- cbind(1,X.tr)
# X2<- cbind(1,X.te)
#
# for(i in 1:11){
# for(j in 1:11){
# w.tr<- cbind(w.tr,X1[,j]*X1[,i])
# w.te<- cbind(w.te, X2[,j]*X2[,i])
# }
# }
#
# B<- matrix(0,ncol = 11,nrow = 11)
# rownames(B)<- c("inter" , paste("X",1:(nrow(B)-1),sep = ""))
# colnames(B)<- c("inter" , paste("X",1:(ncol(B)-1),sep = ""))
#
#
# B_high_SH<- B
# B_high_SH[1:6,1:6]<- 1
# #We exclude quadratic terms in this example
# diag(B_high_SH)[-1]<-0
# #View true coefficient matrix
# pheatmap(as.matrix(B_high_SH), scale="none",
# cluster_rows=FALSE, cluster_cols=FALSE)
#
# #With high Strong heredity: all possible interactions
# Y_high_SH <- as.vector(w.tr%*%as.vector(B_high_SH))+rnorm(100)
# Y_high_SH.te <- as.vector(w.te%*%as.vector(B_high_SH))+rnorm(100)
#
# ############################## FIT SOME MODELS ########################################
#
# #high Strong heredity with l_2 norm
# fit_high_SH<- FAMILY(X.tr, X.tr, Y_high_SH, lambdas ,
# alphas, quad = FALSE,iter=500, verbose = TRUE )
# yhat_hSH<- predict(fit_high_SH, X.te, X.te)
# mse_hSH <-apply(yhat_hSH,c(2,3), "-" ,Y_high_SH.te)
# mse_hSH<- apply(mse_hSH^2,c(2,3),sum)
#
# #Find optimal model and plot matrix
# im<- which(mse_hSH==min(mse_hSH),TRUE)
# plot(fit_high_SH$Estimate[[im[2] ]][[im[1]]])
#
#
# #Plot some matrices for different alpha values
# #Low alpha, higher penalty on groups
# plot(fit_high_SH$Estimate[[ 1 ]][[ 50 ]])
# #Medium alpha, equal penalty on groups and individual interactions
# plot(fit_high_SH$Estimate[[ 2 ]][[ 50 ]])
# #High alpha, more penalty on individual interactions
# plot(fit_high_SH$Estimate[[ 3 ]][[ 50 ]])
#
#
# #View Coefficients
# coef(fit_high_SH,XequalZ = TRUE)[[im[2]]][[im[1]]]
FAMILY documentation built on May 2, 2019, noon
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Description Details Author(s) References See Also Examples
Description
Fits linear and logistic regression models with pairwise interactions for two sets of covariates. The models fitted by this package obey strong heredity which implies that interaction of two covariates is only included if both main effects are included in the model. In the case of high dimensional data we may not wish to model all possible covariates and their interactions but a small subset of them, thus we utilize the penalized regression methods of Haris, Witten and Simon (2014).
Details
| Package: | FAMILY |
| Type: | Package |
| Version: | 0.1.19 |
| Date: | 2015-06-20 |
| License: | GPL-2 |
The package includes the following functions:
FAMILY: | Fit a penalized regression model for a grid of alpha and lambda values |
predict.FAMILY: | Predict yhat or phat for a for a given data set from the fitted model |
coef.FAMILY: | Extract the set of estimated non-zero coefficients |
Author(s)
Asad Haris, Daniela Witten and Noah Simon
Maintainer: Asad Haris <aharis@uw.edu>
References
Haris, Witten and Simon (2014). Convex Modeling of Interactions with Strong Heredity. Available on ArXiv at http://arxiv.org/abs/1410.3517
See Also
FAMILY,
predict.FAMILY,
coef.FAMILY
Examples
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 | library(FAMILY)
library(pROC)
library(pheatmap)
#####################################################################################
#####################################################################################
############################# EXAMPLE - CONTINUOUS RESPONSE #########################
#####################################################################################
#####################################################################################
############################## GENERATE DATA ########################################
#Generate training set of covariates X and Z
set.seed(1)
X.tr<- matrix(rnorm(10*100),ncol = 10, nrow = 100)
Z.tr<- matrix(rnorm(15*100),ncol = 15, nrow = 100)
#Generate test set of covariates X and Z
X.te<- matrix(rnorm(10*100),ncol = 10, nrow = 100)
Z.te<- matrix(rnorm(15*100),ncol = 15, nrow = 100)
#Scale appropiately
meanX<- apply(X.tr,2,mean)
meanY<- apply(Z.tr,2,mean)
X.tr<- scale(X.tr, scale = FALSE)
Z.tr<- scale(Z.tr, scale = FALSE)
X.te<- scale(X.te,center = meanX,scale = FALSE)
Z.te<- scale(Z.te,center = meanY,scale = FALSE)
#Generate full matrix of Covariates
w.tr<- c()
w.te<- c()
X1<- cbind(1,X.tr)
Z1<- cbind(1,Z.tr)
X2<- cbind(1,X.te)
Z2<- cbind(1,Z.te)
for(i in 1:16){
for(j in 1:11){
w.tr<- cbind(w.tr,X1[,j]*Z1[,i])
w.te<- cbind(w.te, X2[,j]*Z2[,i])
}
}
#Generate response variables with signal from
#First 5 X features and 5 Z features.
#We construct the coefficient matrix B.
#B[1,1] contains the intercept
#B[-1,1] contains the main effects for X.
# For instance, B[2,1] is the main effect for the first feature in X.
#B[1,-1] contains the main effects for Z.
# For instance, B[1,10] is the coefficient for the 10th feature in Z.
#B[i+1,j+1] is the coefficient of X_i Z_j
B<- matrix(0,ncol = 16,nrow = 11)
rownames(B)<- c("inter" , paste("X",1:(nrow(B)-1),sep = ""))
colnames(B)<- c("inter" , paste("Z",1:(ncol(B)-1),sep = ""))
# First, we simulate data as follows:
# The first five features in X, and the first five features in Z, are non-zero.
# And given the non-zero main effects, all possible interactions are involved.
# We call this "high strong heredity"
B_high_SH<- B
B_high_SH[1:6,1:6]<- 1
#View true coefficient matrix
pheatmap(as.matrix(B_high_SH), scale="none",
cluster_rows=FALSE, cluster_cols=FALSE)
Y_high_SH <- as.vector(w.tr%*%as.vector(B_high_SH))+rnorm(100,sd = 2)
Y_high_SH.te <- as.vector(w.te%*%as.vector(B_high_SH))+rnorm(100,sd = 2)
# Now a new setting:
# Again, the first five features in X, and the first five features in Z, are involved.
# But this time, only a subset of the possible interactions are involved.
# Strong heredity is still maintained.
# We call this "low strong heredity"
B_low_SH<- B_high_SH
B_low_SH[2:6,2:6]<-0
B_low_SH[3:4,3:5]<- 1
#View true coefficient matrix
pheatmap(as.matrix(B_low_SH), scale="none",
cluster_rows=FALSE, cluster_cols=FALSE)
Y_low_SH <- as.vector(w.tr%*%as.vector(B_low_SH))+rnorm(100,sd = 1.5)
Y_low_SH.te <- as.vector(w.te%*%as.vector(B_low_SH))+rnorm(100,sd = 1.5)
############################## FIT SOME MODELS ########################################
#Define alphas and lambdas
#Define 3 different alpha values
#Low alpha values penalize groups more
#High alpha values penalize individual Interactions more
alphas<- c(0.01,0.5,0.99)
lambdas<- seq(0.1,1,length = 50)
#high Strong heredity with l2 norm
fit_high_SH<- FAMILY(X.tr, Z.tr, Y_high_SH, lambdas ,
alphas, quad = TRUE,iter=500, verbose = TRUE )
yhat_hSH<- predict(fit_high_SH, X.te, Z.te)
mse_hSH <-apply(yhat_hSH,c(2,3), "-" ,Y_high_SH.te)
mse_hSH<- apply(mse_hSH^2,c(2,3),sum)
#Find optimal model and plot matrix
im<- which(mse_hSH==min(mse_hSH),TRUE)
plot(fit_high_SH$Estimate[[im[2] ]][[im[1]]])
#Plot some matrices for different alpha values
#Low alpha, higher penalty on groups
plot(fit_high_SH$Estimate[[ 1 ]][[ 25 ]])
#Medium alpha, equal penalty on groups and individual interactions
plot(fit_high_SH$Estimate[[ 2 ]][[ 25 ]])
#High alpha, more penalty on individual interactions
plot(fit_high_SH$Estimate[[ 3 ]][[ 40 ]])
#View Coefficients
coef(fit_high_SH)[[im[2]]][[im[1]]]
############################## Uncomment code for EXAMPLE ###########################
# #high Strong heredity with l_infinity norm norm
# fit_high_SH<- FAMILY(X.tr, Z.tr, Y_high_SH, lambdas ,
# alphas, quad = TRUE,iter=500, verbose = TRUE,
# norm = "l_inf")
# yhat_hSH<- predict(fit_high_SH, X.te, Z.te)
# mse_hSH <-apply(yhat_hSH,c(2,3), "-" ,Y_high_SH.te)
# mse_hSH<- apply(mse_hSH^2,c(2,3),sum)
#
# #Find optimal model and plot matrix
# im<- which(mse_hSH==min(mse_hSH),TRUE)
# plot(fit_high_SH$Estimate[[im[2] ]][[im[1]]])
#
#
# #Plot some matrices for different alpha values
# #Low alpha, higher penalty on groups
# plot(fit_high_SH$Estimate[[ 1 ]][[ 30 ]])
# #Medium alpha, equal penalty on groups and individual interactions
# plot(fit_high_SH$Estimate[[ 2 ]][[ 10 ]])
# #High alpha, more penalty on individual interactions
# plot(fit_high_SH$Estimate[[ 3 ]][[ 20 ]])
#
#
# #View Coefficients
# coef(fit_high_SH)[[im[2]]][[im[1]]]
############################## Uncomment code for EXAMPLE ###########################
# #Redefine lambdas
# lambdas<- seq(0.1,0.5,length = 50)
#
# #low Strong heredity with l_2 norm
# fit_low_SH<- FAMILY(X.tr, Z.tr, Y_low_SH, lambdas ,
# alphas, quad = TRUE,iter=500, verbose = TRUE )
# yhat_lSH<- predict(fit_low_SH, X.te, Z.te)
# mse_lSH <-apply(yhat_lSH,c(2,3), "-" ,Y_low_SH.te)
# mse_lSH<- apply(mse_lSH^2,c(2,3),sum)
#
# #Find optimal model and plot matrix
# im<- which(mse_lSH==min(mse_lSH),TRUE)
# plot(fit_low_SH$Estimate[[im[2] ]][[im[1]]])
#
#
# #Plot some matrices for different alpha values
# #Low alpha, higher penalty on groups
# plot(fit_low_SH$Estimate[[ 1 ]][[ 25 ]])
# #Medium alpha, equal penalty on groups and individual interactions
# plot(fit_low_SH$Estimate[[ 2 ]][[ 10 ]])
# #High alpha, more penalty on individual interactions
# plot(fit_low_SH$Estimate[[ 3 ]][[ 10 ]])
#
#
# #View Coefficients
# coef(fit_low_SH)[[im[2]]][[im[1]]]
#####################################################################################
#####################################################################################
############################### EXAMPLE - BINARY RESPONSE ###########################
#####################################################################################
#####################################################################################
############################## GENERATE DATA ########################################
#Generate data for logistic regression
Yp_high_SH<- as.vector((w.tr)%*%as.vector(B_high_SH))
Yp_high_SH.te<- as.vector((w.te)%*%as.vector(B_high_SH))
Yprobs_high_SH<- 1/(1+exp(-Yp_high_SH))
Yprobs_high_SH.te<- 1/(1+exp(-Yp_high_SH.te))
Yp_high_SH<- rbinom(100, size = 1, prob = Yprobs_high_SH)
Yp_high_SH.te<- rbinom(100, size = 1, prob = Yprobs_high_SH.te)
lambdas<- seq(0.01,0.15,length = 50)
############################## FIT SOME MODELS ########################################
#Fit glm via l_2 norm
fit_high_SH<- FAMILY(X.tr, Z.tr, Yp_high_SH, lambdas ,
alphas, quad = TRUE,iter=500, verbose = TRUE,
family = "binomial")
yhp_hSH<- predict(fit_high_SH, X.te, Z.te)
mse_high_SH <-apply(yhp_hSH,c(2,3), "-" ,Yp_high_SH.te)
mse_hSH<- apply(mse_high_SH^2,c(2,3),sum)
im<- which(mse_hSH==min(mse_hSH),TRUE)
plot(fit_high_SH$Estimate[[im[2] ]][[im[1]]])
roc( Yp_high_SH.te,yhp_hSH[,im[1],im[2]],plot = TRUE)
#View Coefficients
coef(fit_high_SH)[[im[2]]][[im[1]]]
############################## Uncomment code for EXAMPLE ###########################
# #Fit glm via l_infinity norm
# fit_high_SH<- FAMILY(X.tr, Z.tr, Yp_high_SH, lambdas , norm = "l_inf",
# alphas, quad = TRUE,iter=500, verbose = TRUE,
# family = "binomial")
# yhp_hSH<- predict(fit_high_SH, X.te, Z.te)
# mse_high_SH <-apply(yhp_hSH,c(2,3), "-" ,Yp_high_SH.te)
# mse_hSH<- apply(mse_high_SH^2,c(2,3),sum)
# im<- which(mse_hSH==min(mse_hSH),TRUE)
# plot(fit_high_SH$Estimate[[im[2] ]][[im[1]]])
# roc( Yp_high_SH.te,yhp_hSH[,im[1],im[2]],plot = TRUE)
#
# #View Coefficients
# coef(fit_high_SH)[[im[2]]][[im[1]]]
#####################################################################################
#####################################################################################
############################## EXAMPLE WHERE X=Z ####################################
######################## Uncomment Code for EXAMPLE #################################
#####################################################################################
############################## GENERATE DATA ########################################
# #Redefine Lambdas
# lambdas<- seq(0.01,0.3,length = 50)
#
#
# #We consider the case X=Z now
# w.tr<- c()
# w.te<- c()
# X1<- cbind(1,X.tr)
# X2<- cbind(1,X.te)
#
# for(i in 1:11){
# for(j in 1:11){
# w.tr<- cbind(w.tr,X1[,j]*X1[,i])
# w.te<- cbind(w.te, X2[,j]*X2[,i])
# }
# }
#
# B<- matrix(0,ncol = 11,nrow = 11)
# rownames(B)<- c("inter" , paste("X",1:(nrow(B)-1),sep = ""))
# colnames(B)<- c("inter" , paste("X",1:(ncol(B)-1),sep = ""))
#
#
# B_high_SH<- B
# B_high_SH[1:6,1:6]<- 1
# #We exclude quadratic terms in this example
# diag(B_high_SH)[-1]<-0
# #View true coefficient matrix
# pheatmap(as.matrix(B_high_SH), scale="none",
# cluster_rows=FALSE, cluster_cols=FALSE)
#
# #With high Strong heredity: all possible interactions
# Y_high_SH <- as.vector(w.tr%*%as.vector(B_high_SH))+rnorm(100)
# Y_high_SH.te <- as.vector(w.te%*%as.vector(B_high_SH))+rnorm(100)
#
# ############################## FIT SOME MODELS ########################################
#
# #high Strong heredity with l_2 norm
# fit_high_SH<- FAMILY(X.tr, X.tr, Y_high_SH, lambdas ,
# alphas, quad = FALSE,iter=500, verbose = TRUE )
# yhat_hSH<- predict(fit_high_SH, X.te, X.te)
# mse_hSH <-apply(yhat_hSH,c(2,3), "-" ,Y_high_SH.te)
# mse_hSH<- apply(mse_hSH^2,c(2,3),sum)
#
# #Find optimal model and plot matrix
# im<- which(mse_hSH==min(mse_hSH),TRUE)
# plot(fit_high_SH$Estimate[[im[2] ]][[im[1]]])
#
#
# #Plot some matrices for different alpha values
# #Low alpha, higher penalty on groups
# plot(fit_high_SH$Estimate[[ 1 ]][[ 50 ]])
# #Medium alpha, equal penalty on groups and individual interactions
# plot(fit_high_SH$Estimate[[ 2 ]][[ 50 ]])
# #High alpha, more penalty on individual interactions
# plot(fit_high_SH$Estimate[[ 3 ]][[ 50 ]])
#
#
# #View Coefficients
# coef(fit_high_SH,XequalZ = TRUE)[[im[2]]][[im[1]]]
|
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.