cv.FuncompCGL: Cross-validation for FuncompCGL
In jiji6454/Rpac_compReg: Compositional Data Regression
Description
Usage
Arguments
Value
Examples
Description
Does nfolds cross-validation for FuncompCGL, return value of lam
and df k. The function is modified based on the cv function
from glmnet package
Usage
1
2
3
4
5
6
Arguments
y
a vector of response variable.
X
a data frame of longitudinal compositinal predictors with number p,
subject ID and time variable. Order of subject ID should be the same as that of y.
If df k is a scalar, X could be the matrix with dimension n*(k*p - length(ref))
after integral.
Zc
A design matrix for control variables, could be missing. Default is NULL. No penalty is imposed.
lam
a user supplied lambda sequence. Typically, by leaving this option unspecified users can have the
program compute its own lam sequence based on nlam and lambda.factor
If lam is provided but a scaler, lam sequence is also created starting from lam.
Supplying a value of lambda overrides this. It is better to supply a decreasing sequence of lambda
values, if not, the program will sort user-defined lambda sequence in decreasing order
automatically.
ref
reference variable. If ref is set to a scalar between [1,p], log-contract method is applied with the variable
ref as baseline. If ref = NULL (default value), constrained group lasso method is applied
W
a vector in length of p (the total number of groups), matrix with dimension p1*p1 or character specifying function
used to calculate inverted weight matrix for each group.
If vector, works as penalty factor. Separate penalty weights can be applied to each group of beta'ss.
to allow differential shrinkage. Can be 0 for some groups, which implies no shrinkage, and results in that group
always being included in the model.
If matrix, a block diagonal matrix. Diagonal elements are inverted weights matrics for each group.
if character, user should provide the function for inverted weights matrics.
Default value is rep(1, times = p).
k
a vector or scalar consists of df for basis - default is 4:10.
foldid
an optional vector of values between 1 and nfolds identifying what fold each
observation is in. If supplied, nfold can be missing.
nfolds
number of folds - default is 10. Smallest value allowable is nfolds=3.
nlam
the length of lam sequence - default is 100.
trim
a scaler specifying percentage to be trimmed off for prediction error - default is 0.
This feature could be deleted later.
outer_maxiter
maximun munber of loops allowed for Augmented Lanrange method.
keep
If keep=TRUE, a prevalidated array is returned containing fitted values for each observation,
of lambda and k. This means these fits are computed with this observation and the rest of
its fold omitted. The folid vector is also returned. Default is keep=FALSE
...
other arguments that can be passed to FuncompCGL.
Value
a list
Funcomp.CGL.fit
a list, length of k,
of fitted FuncompCGL object for the full data.
objects with S3 calss FuncompCGL
lam
the values of lam used in the fits
Ftrim
a list for cross-validation result with trim = 0.
-
cvm the mean cross-validated error without trimming - a matrix of length(k)*length(lam)
-
cvsd estimate of standard error of cvm without trimming- a matrix of length(k)*length(lam)
-
cvup a matrix of length(k)*length(lam)
-
cvlo a matrix of length(k)*length(lam)
-
lam.min The optimal value of k and lam that gives minimum cross validation error cvm without trimming
-
lam.1se The optimal value of k and lam for "lam.1se" without trimming
Ttrim
a list for cross-validation result with trimming (if trim provided).
Same as these for Ftrim.
This feature could be deleted later.
fit.preval, foldid
fitting matrix and foldid. Only keept when keep=TRUE.
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
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
df_beta = 5
p = 30
beta_C_true = matrix(0, nrow = p, ncol = df_beta)
beta_C_true[3, ] <- c(-1, 0, 0, 0, -0.5)
beta_C_true[1, ] <- c(1, 0, 1 , 0, -0.5)
beta_C_true[2, ] <- c(0, 0, -1, 0, 1)
nfolds = 10
k_list <- c(4,5)
n_train = 100
n_test = 500
Data <- Model2(n = 100, p = p, m = 0, intercept = TRUE,
SNR = 2, sigma = 2,
rho_X = 0, rho_W = 0.5,
Corr_X = "CorrCS", Corr_W = "CorrAR",
df_W = 5, df_beta = df_beta,
ns = 20, obs_spar = 1, theta.add = FALSE, #c(0,0,0),
beta_C = as.vector(t(beta_C_true)))
y <- drop(Data$data$y)
n <- length(y)
X <- Data$data$Comp
Zc <- Data$data$Zc
intercept <- Data$data$intercept
m <- ifelse(is.null(Zc), 0, dim(Zc)[2]) #+ as.integer(intercept)
m1 <- m + as.integer(intercept)
sseq <- Data$basis.info[,1]
beta_C.true <- matrix(Data$beta[1:(p*(df_beta))],
nrow = p, ncol = df_beta, byrow = TRUE)
beta_curve.true <- Data$basis.info[,-1] %*% t(beta_C.true)
Non_zero.true <- (1:p)[apply(beta_C.true, 1, function(x) max(abs(x)) > 0)]
foldid <- sample(rep(seq(nfolds), length = n))
arg_list <- as.list(Data$call)[-1]
arg_list$n <- n_test
Test <- do.call(Model2, arg_list)
y_test <- drop(Test$data$y)
# y = y
# X = X
# Zc = Zc
# intercept = intercept
# W = rep(1, p)
# k = k_list
# nfolds = 10
# trim = 0
# tol = 0
# inner_eps = 1e-6
# inner_maxiter = 1E3
# dfmax = 20
# lambda.factor = 1e-20
# mu_ratio = 1
# outer_eps = 1e-6
# keep = TRUE
# Trange = c(0,1)
rule <- "lam.min"
# cv_cgl, Constrained group lasso
cv_cgl <- cv.FuncompCGL(y = y, X = X, Zc = Zc, intercept = intercept,
W = rep(1, p), #W = function(x){ diag( 1 / apply(x, 2, sd) ) },
k = k_list, trim = 0,
foldid = foldid, #nfolds = 10,
tol = 0, inner_eps = 1e-6, inner_maxiter = 1E3,
dfmax = 30, lambda.factor = 1e-3,
mu_ratio = 1, outer_eps = 1e-6,
keep = TRUE, Trange = c(0,1)
#lam = c(exp(seq(log(0.01),log(0.00001),length=100)),0)
)
plot(cv_cgl,k_list = k_list)
cv_cgl$Ftrim[c("lam.min", "lam.1se")]
beta <- coef(cv_cgl, trim = FALSE, s = rule)
k_opt <- (length(drop(beta)) - (m + 1)) / p
#cv_cgl$Funcomp.CGL.fit[[as.character(k_opt)]]
par(mfrow=c(1,4))
plot(cv_cgl,k_list = k_list)
plot(cv_cgl$Funcomp.CGL.fit[[as.character(k_opt)]],
p = p, k = k_opt, ylab = "L2")
plot(cv_cgl$Ftrim$cvm[k_opt - k_list[1] + 1, ])
title(paste0("k=", k_opt), line = 0.5)
# apply(cv_cgl$Funcomp.CGL.fit[[k_opt - 3]]$beta, 2,
# function(x, p, k) {
# #which(abs(x[seq(1, (p-1)*k+1, by = k)])>0),
# (1:p)[apply(matrix(x[1:(p*k)], byrow = TRUE, nrow = p), 1,
# function(x) max(abs(x)) > 0)]
# },p = p , k = k_opt)
if(k_opt == df_beta) {
plot(Data$beta, col = "red", pch = 19,
ylim = range(c(range(Data$beta), range(beta)))) #range(Data$beta))
abline(v= seq(from = 0, to = (p*df_beta), by = df_beta ))
abline(h = 0)
points(beta)
if(m1 > 0) points(p*df_beta + 1:m1, tail(Data$beta, m1),
col = "blue", pch = 19)
} else {
plot(beta, ylim = range(c(range(Data$beta), range(beta))) )
abline(v= seq(from = 0, to = (p*k_opt), by = k_opt ))
abline(h = 0, col = "red")
if(m1 > 0) points(p*k_opt + 1:m1, tail(Data$beta, m1),
col = "blue", pch = 19)
}
title(paste0("Method cgl"), outer=TRUE, line = -2)
par(mfrow=c(1,1))
beta_C <- matrix(beta[1:(p*k_opt)], byrow = TRUE, nrow = p)
cat("colSums:", colSums(beta_C))
#Non.zero <- which(abs(beta_C[,1]) > 0)
Non.zero <- apply(beta_C, 1, function(x) ifelse(max(abs(x)) >0, TRUE, FALSE))
Non.zero <- (1:p)[Non.zero]
cat("None zero groups:", Non.zero)
#vet(beta, p = p, k = k_opt)
par(mfrow=c(1,4))
plot(cv_cgl) #plot(cv_cgl,k_list = k_list)
matplot(sseq, beta_curve.true,
ylab = "coeffcients curve", xlab = "TIME", #main = "TRUE",
ylim = range(Data$beta[1:(p*df_beta)]),
type = "l")
abline(a = 0, b = 0, col = "grey", lwd = 2)
title("TRUE", line = 0.5)
text(0, beta_curve.true[1, Non_zero.true], labels = paste(Non_zero.true))
B <- splines::bs(Data$basis.info[,1], df = k_opt, intercept = TRUE)
beta_curve <- B %*% t(beta_C)
matplot(sseq, beta_curve,
ylab = "coef", xlab = "TIME", #main = "ESTI",
ylim = range(Data$beta[1:(p*df_beta)])#,
#type = "l"
)
abline(a = 0, b = 0, col = "grey", lwd = 2)
title("Estimate", line = 0.5)
text(0, beta_curve[1, Non.zero], labels = paste(Non.zero))
text(tail(sseq, 1), beta_curve[dim(beta_curve)[1], Non.zero], labels = paste(Non.zero))
plot(apply(abs(beta_C),1,sum))
title(paste0("k=", k_opt), line = 0.5)
title(paste0("Method cgl"), outer=TRUE, line = -2)
par(mfrow=c(1,1))
##set a cutoff when you compute nonzeros
Non.zero <- apply(beta_C, 1, function(x)
ifelse(sqrt(sum(x^2)) > sqrt(sum(beta_C^2))/100, TRUE, FALSE))
Non.zero <- (1:p)[Non.zero]
Non.zero
Non.zero <- apply(beta_C, 1, function(x)
ifelse(sum(x^2) > sum(beta_C^2)/100, TRUE, FALSE))
Non.zero <- (1:p)[Non.zero]
Non.zero
## cut by curve
Curve_L2 <- colSums(beta_curve^2)
Curve_L2 <- Curve_L2 - colSums(beta_curve[c(1, nrow(Data$basis.info)), ]^2) / 2
Curve_L2 <- Curve_L2 * (Data$basis.info[2,1] - Data$basis.info[1,1])
plot(Curve_L2)
which(sqrt(Curve_L2) > sqrt(sum(Curve_L2)) / 100)
cgl <- list()
# MSE <- crossprod(y - cbind2(cbind(cv_cgl$Funcomp.CGL.fit[[k_opt - k_list[1]+ 1]]$Z,
# Zc), 1) %*% beta) / length(y)
X_train <- cbind2(cbind(cv_cgl$Funcomp.CGL.fit[[k_opt - k_list[1]+ 1]]$Z, Zc), 1)
MSE <- sum((y - X_train %*% beta)^2) / length(y)
# R_sqr <- 1 - crossprod(y - cbind2(cbind(cv_cgl$Funcomp.CGL.fit[[k_opt - k_list[1]+ 1]]$Z,
# Zc), 1) %*% beta) / crossprod(y - mean(y))
R_sqr <- sum((y - X_train %*% beta)^2)
R_sqr <- 1 - R_sqr / crossprod(y - mean(y))
obj <- FuncompCGL(y = y_test, X = Test$data$Comp, k = k_opt, nlam = 1, outer_maxiter = 0)
# PE <- sum((Test$data$y - cbind2(cbind(obj$Z, Test$data$Zc), 1) %*% beta)^2)
# / length(drop(Test$data$y))
X_test <- cbind2(cbind(obj$Z, Test$data$Zc), 1)
PE <- sum((y_test - X_test %*% beta)^2) / length(y_test)
cgl$pred_error <- c(MSE = MSE, PE = PE, Rsqr_train = R_sqr)
cgl$Non.zero_cut <- Non.zero
cgl <- c(cgl,
ERROR_fun(beta_fit = beta, beta_true = Data$beta,
basis_fit = B, basis_true = Data$basis.info[,-1],
sseq = Data$basis.info[, 1],
m = m, p = p, Nzero_group = length(Non_zero.true), tol = 0),
k = k_opt)
cgl$coef <- list(beta_C = beta_C, beta_c = tail(beta, m1))
## Not run:
# naive model
# set mu_raio = 0 to identifying without linear constraints,
# no outer_loop for Lagrange augmented multiplier
# mu_ratio = 0
cv_naive <- cv.FuncompCGL(y = y, X = X, Zc = Zc, intercept = intercept,
W = rep(1, p), #W = function(x){ diag( 1 / apply(x, 2, sd) ) },
k = k_list, nfolds = 10, trim = 0,
tol = 0, inner_eps = 1e-6, inner_maxiter = 1E3,
dfmax = 30, lambda.factor = 1e-3,
mu_ratio = 0, outer_eps = 1e-6,
keep = FALSE, Trange = c(0,1)
#lam = c(exp(seq(log(0.01),log(0.00001),length=100)),0)
)
plot(cv_naive,k_list = k_list)
cv_naive$Ftrim[c("lam.min", "lam.1se")]
beta <- coef(cv_naive, trim = FALSE, s = rule)
k_opt <- (length(drop(beta)) - (m + 1)) / p
#cv_naive$Funcomp.CGL.fit[[as.character(k_opt)]]
par(mfrow=c(1,4))
plot(cv_naive,k_list = k_list)
plot(cv_naive$Funcomp.CGL.fit[[as.character(k_opt)]],
p = p, k = k_opt, ylab = "L2")
plot(cv_naive$Ftrim$cvm[k_opt - k_list[1] + 1, ])
title(paste0("k=", k_opt), line = 0.5)
# apply(cv_naive$Funcomp.CGL.fit[[k_opt - 3]]$beta, 2,
# function(x, p, k) {
# #which(abs(x[seq(1, (p-1)*k+1, by = k)])>0),
# (1:p)[apply(matrix(x[1:(p*k)], byrow = TRUE, nrow = p), 1,
# function(x) max(abs(x)) > 0)]
# },p = p , k = k_opt)
if(k_opt == df_beta) {
plot(Data$beta, col = "red", pch = 19,
ylim = range(c(range(Data$beta), range(beta)))) #range(Data$beta))
abline(v= seq(from = 0, to = (p*df_beta), by = df_beta ))
abline(h = 0)
points(beta)
if(m1 > 0) points(p*df_beta + 1:m1, tail(Data$beta, m1),
col = "blue", pch = 19)
} else {
plot(beta, ylim = range(c(range(Data$beta), range(beta))) )
abline(v= seq(from = 0, to = (p*k_opt), by = k_opt ))
abline(h = 0, col = "red")
if(m1 > 0) points(p*k_opt + 1:m1, tail(Data$beta, m1),
col = "blue", pch = 19)
}
title(paste0("Method naive"), outer=TRUE, line = -2)
par(mfrow=c(1,1))
beta_C <- matrix(beta[1:(p*k_opt)], byrow = TRUE, nrow = p)
cat("colSums:", colSums(beta_C))
#Non.zero <- which(abs(beta_C[,1]) > 0)
Non.zero <- apply(beta_C, 1, function(x) ifelse(max(abs(x)) >0, TRUE, FALSE))
Non.zero <- (1:p)[Non.zero]
cat("None zero groups:", Non.zero)
#vet(beta, p = p, k = k_opt)
par(mfrow=c(1,4))
plot(cv_naive) #plot(cv_naive,k_list = k_list)
matplot(sseq, beta_curve.true,
ylab = "coeffcients curve", xlab = "TIME", #main = "TRUE",
ylim = range(Data$beta[1:(p*df_beta)]),
type = "l")
abline(a = 0, b = 0, col = "grey", lwd = 2)
title("TRUE", line = 0.5)
text(0, beta_curve.true[1, Non_zero.true], labels = paste(Non_zero.true))
B <- splines::bs(Data$basis.info[,1], df = k_opt, intercept = TRUE)
beta_curve <- B %*% t(beta_C)
matplot(sseq, beta_curve,
ylab = "coef", xlab = "TIME", #main = "ESTI",
ylim = range(Data$beta[1:(p*df_beta)])#,
#type = "l"
)
abline(a = 0, b = 0, col = "grey", lwd = 2)
title("Estimate", line = 0.5)
text(0, beta_curve[1, Non.zero], labels = paste(Non.zero))
text(tail(sseq, 1), beta_curve[dim(beta_curve)[1], Non.zero], labels = paste(Non.zero))
plot(apply(abs(beta_C),1,sum))
title(paste0("k=", k_opt), line = 0.5)
title(paste0("Method naive"), outer=TRUE, line = -2)
par(mfrow=c(1,1))
##set a cutoff when you compute nonzeros
Non.zero <- apply(beta_C, 1, function(x)
ifelse(sqrt(sum(x^2)) > sqrt(sum(beta_C^2))/100, TRUE, FALSE))
Non.zero <- (1:p)[Non.zero]
Non.zero
naive <- list()
# MSE <- crossprod(y - cbind2(cbind(cv_naive$Funcomp.CGL.fit[[k_opt - k_list[1]+ 1]]$Z,
# Zc), 1) %*% beta)
X_train <- cbind2(cbind(cv_naive$Funcomp.CGL.fit[[k_opt - k_list[1]+ 1]]$Z, Zc), 1)
MSE <- sum((y - X_train %*% beta)^2)/ length(y)
# R_sqr <- 1 - crossprod(y - cbind2(cbind(cv_naive$Funcomp.CGL.fit[[k_opt - k_list[1]+ 1]]$Z,
# Zc), 1) %*% beta) / crossprod(y - mean(y))
R_sqr <- sum((y - X_train %*% beta)^2)
R_sqr <- 1 - R_sqr / crossprod(y - mean(y))
obj <- FuncompCGL(y = Test$data$y, X = Test$data$Comp, k = k_opt, nlam = 1, outer_maxiter = 0)
# PE <- sum((Test$data$y - cbind2(cbind(obj$Z, Test$data$Zc), 1) %*% beta)^2)
# / length(drop(Test$data$y))
X_test <- cbind2(cbind(obj$Z, Test$data$Zc), 1)
PE <- sum((y_test - X_test %*% beta)^2) / length(y_test)
naive$pred_error <- c(MSE = MSE, PE = PE, Rsqr_train = R_sqr)
naive$Non.zero_cut <- Non.zero
naive <- c(naive,
ERROR_fun(beta_fit = beta, beta_true = Data$beta,
basis_fit = B, basis_true = Data$basis.info[,-1],
sseq = Data$basis.info[, 1],
m = m, p = p, Nzero_group = length(Non_zero.true), tol = 0),
k = k_opt)
naive$coef <- list(beta_C = beta_C, beta_c = tail(beta, m1))
# log contract model
# set reference variable and once ref is set to a scalar in range,
# mu_ratio is set to 0 automatically
# ref = sample(1:p, 1)
cv_base <- cv.FuncompCGL(y = y, X = X, Zc = Zc, intercept = intercept, ref = sample(1:p, 1),
W = rep(1, p - 1), #W = function(x){ diag( 1 / apply(x, 2, sd) ) },
k = k_list, nfolds = 10, trim = 0,
tol = 0, inner_eps = 1e-6, inner_maxiter = 1E3,
dfmax = 30, lambda.factor = 1e-3,
mu_ratio = 0, outer_eps = 1e-6,
keep = FALSE, Trange = c(0,1)
#,lam = c(exp(seq(log(0.01),log(0.00001),length=100)),0)
)
plot(cv_base,k_list = k_list)
cv_base$Ftrim[c("lam.min", "lam.1se")]
beta <- coef(cv_base, trim = FALSE, s = rule)
k_opt <- (length(drop(beta)) - (m + 1)) / p
#cv_base$Funcomp.CGL.fit[[as.character(k_opt)]]
par(mfrow=c(1,4))
plot(cv_base,k_list = k_list)
plot(cv_base$Funcomp.CGL.fit[[as.character(k_opt)]],
p = p, k = k_opt, ylab = "L2")
plot(cv_base$Ftrim$cvm[k_opt - k_list[1] + 1, ])
title(paste0("k=", k_opt), line = 0.5)
# apply(cv_base$Funcomp.CGL.fit[[k_opt - 3]]$beta, 2,
# function(x, p, k) {
# #which(abs(x[seq(1, (p-1)*k+1, by = k)])>0),
# (1:p)[apply(matrix(x[1:(p*k)], byrow = TRUE, nrow = p), 1,
# function(x) max(abs(x)) > 0)]
# },p = p , k = k_opt)
if(k_opt == df_beta) {
plot(Data$beta, col = "red", pch = 19,
ylim = range(c(range(Data$beta), range(beta)))) #range(Data$beta))
abline(v= seq(from = 0, to = (p*df_beta), by = df_beta ))
abline(h = 0)
points(beta)
if(m1 > 0) points(p*df_beta + 1:m1, tail(Data$beta, m1),
col = "blue", pch = 19)
} else {
plot(beta, ylim = range(c(range(Data$beta), range(beta))) )
abline(v= seq(from = 0, to = (p*k_opt), by = k_opt ))
abline(h = 0, col = "red")
if(m1 > 0) points(p*k_opt + 1:m1, tail(Data$beta, m1),
col = "blue", pch = 19)
}
title(paste0("Method baseline=", cv_base$Funcomp.CGL.fit[[1]]$ref), outer=TRUE, line = -2)
par(mfrow=c(1,1))
beta_C <- matrix(beta[1:(p*k_opt)], byrow = TRUE, nrow = p)
cat("colSums:", colSums(beta_C))
#Non.zero <- which(abs(beta_C[,1]) > 0)
Non.zero <- apply(beta_C, 1, function(x) ifelse(max(abs(x)) >0, TRUE, FALSE))
Non.zero <- (1:p)[Non.zero]
cat("None zero groups:", Non.zero)
#vet(beta, p = p, k = k_opt)
par(mfrow=c(1,4))
plot(cv_base) #plot(cv_base,k_list = k_list)
matplot(sseq, beta_curve.true,
ylab = "coeffcients curve", xlab = "TIME", #main = "TRUE",
ylim = range(Data$beta[1:(p*df_beta)]),
type = "l")
abline(a = 0, b = 0, col = "grey", lwd = 2)
title("TRUE", line = 0.5)
text(0, beta_curve.true[1, Non_zero.true], labels = paste(Non_zero.true))
B <- splines::bs(Data$basis.info[,1], df = k_opt, intercept = TRUE)
beta_curve <- B %*% t(beta_C)
matplot(sseq, beta_curve,
ylab = "coef", xlab = "TIME", #main = "ESTI",
ylim = range(Data$beta[1:(p*df_beta)])#,
#type = "l"
)
abline(a = 0, b = 0, col = "grey", lwd = 2)
title("Estimate", line = 0.5)
text(0, beta_curve[1, Non.zero], labels = paste(Non.zero))
text(tail(sseq, 1), beta_curve[dim(beta_curve)[1], Non.zero], labels = paste(Non.zero))
plot(apply(abs(beta_C),1,sum))
title(paste0("k=", k_opt), line = 0.5)
title(paste0("Method baseline=", cv_base$Funcomp.CGL.fit[[1]]$ref), outer=TRUE, line = -2)
par(mfrow=c(1,1))
##set a cutoff when you compute nonzeros
Non.zero <- apply(beta_C, 1, function(x)
ifelse(sqrt(sum(x^2)) > sqrt(sum(beta_C^2))/100, TRUE, FALSE))
Non.zero <- (1:p)[Non.zero]
Non.zero
base <- list()
# MSE <- crossprod(y - cbind2(cbind(cv_cgl$Funcomp.CGL.fit[[k_opt - k_list[1]+ 1]]$Z,
# Zc), 1) %*% beta) / length(y)
X_train <- cbind2(cbind(cv_cgl$Funcomp.CGL.fit[[k_opt - k_list[1]+ 1]]$Z, Zc), 1)
MSE <- crossprod(y - X_train %*% beta) / length(y)
# R_sqr <- 1 - crossprod(y - cbind2(cbind(cv_cgl$Funcomp.CGL.fit[[k_opt - k_list[1]+ 1]]$Z,
# Zc), 1) %*% beta) / crossprod(y - mean(y))
R_sqr <- crossprod(y - X_train %*% beta)
R_sqr <- 1 - R_sqr / crossprod(y - mean(y))
obj <- FuncompCGL(y = Test$data$y, X = Test$data$Comp, k = k_opt, nlam = 1, outer_maxiter = 0)
# PE <- sum((Test$data$y - cbind2(cbind(obj$Z, Test$data$Zc), 1) %*% beta)^2)
# / length(drop(Test$data$y))
X_test <- cbind2(cbind(obj$Z, Test$data$Zc), 1)
PE <- sum((y_test - X_test %*% beta)^2) / length(y_test)
base$pred_error <- c(MSE = MSE, PE = PE, Rsqr_train = R_sqr)
base$Non.zero_cut <- Non.zero
base <- c(base,
ERROR_fun(beta_fit = beta, beta_true = Data$beta,
basis_fit = B, basis_true = Data$basis.info[,-1],
sseq = Data$basis.info[, 1],
m = m, p = p, Nzero_group = length(Non_zero.true), tol = 0),
k = k_opt)
base$coef <- list(beta_C = beta_C, beta_c = tail(beta, m1))
## End(Not run)
jiji6454/Rpac_compReg documentation built on May 31, 2019, 5:01 a.m.
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 Usage Arguments Value Examples
Description
Does nfolds cross-validation for FuncompCGL, return value of lam
and df k. The function is modified based on the cv function
from glmnet package
Usage
1 2 3 4 5 6 |
Arguments
y |
a vector of response variable. |
X |
a data frame of longitudinal compositinal predictors with number p,
subject ID and time variable. Order of subject ID should be the same as that of |
Zc |
A design matrix for control variables, could be missing. Default is NULL. No penalty is imposed. |
lam |
a user supplied lambda sequence. Typically, by leaving this option unspecified users can have the
program compute its own |
ref |
reference variable. If |
W |
a vector in length of p (the total number of groups), matrix with dimension
Default value is rep(1, times = p). |
k |
a vector or scalar consists of df for basis - default is 4:10. |
foldid |
an optional vector of values between 1 and |
nfolds |
number of folds - default is 10. Smallest value allowable is nfolds=3. |
nlam |
the length of |
trim |
a scaler specifying percentage to be trimmed off for prediction error - default is 0. |
outer_maxiter |
maximun munber of loops allowed for Augmented Lanrange method. |
keep |
If |
... |
other arguments that can be passed to |
Value
a list
Funcomp.CGL.fit |
a list, length of |
lam |
the values of |
Ftrim |
a list for cross-validation result with trim = 0.
|
Ttrim |
a list for cross-validation result with trimming (if |
fit.preval, foldid |
fitting matrix and foldid. Only keept when |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 | df_beta = 5
p = 30
beta_C_true = matrix(0, nrow = p, ncol = df_beta)
beta_C_true[3, ] <- c(-1, 0, 0, 0, -0.5)
beta_C_true[1, ] <- c(1, 0, 1 , 0, -0.5)
beta_C_true[2, ] <- c(0, 0, -1, 0, 1)
nfolds = 10
k_list <- c(4,5)
n_train = 100
n_test = 500
Data <- Model2(n = 100, p = p, m = 0, intercept = TRUE,
SNR = 2, sigma = 2,
rho_X = 0, rho_W = 0.5,
Corr_X = "CorrCS", Corr_W = "CorrAR",
df_W = 5, df_beta = df_beta,
ns = 20, obs_spar = 1, theta.add = FALSE, #c(0,0,0),
beta_C = as.vector(t(beta_C_true)))
y <- drop(Data$data$y)
n <- length(y)
X <- Data$data$Comp
Zc <- Data$data$Zc
intercept <- Data$data$intercept
m <- ifelse(is.null(Zc), 0, dim(Zc)[2]) #+ as.integer(intercept)
m1 <- m + as.integer(intercept)
sseq <- Data$basis.info[,1]
beta_C.true <- matrix(Data$beta[1:(p*(df_beta))],
nrow = p, ncol = df_beta, byrow = TRUE)
beta_curve.true <- Data$basis.info[,-1] %*% t(beta_C.true)
Non_zero.true <- (1:p)[apply(beta_C.true, 1, function(x) max(abs(x)) > 0)]
foldid <- sample(rep(seq(nfolds), length = n))
arg_list <- as.list(Data$call)[-1]
arg_list$n <- n_test
Test <- do.call(Model2, arg_list)
y_test <- drop(Test$data$y)
# y = y
# X = X
# Zc = Zc
# intercept = intercept
# W = rep(1, p)
# k = k_list
# nfolds = 10
# trim = 0
# tol = 0
# inner_eps = 1e-6
# inner_maxiter = 1E3
# dfmax = 20
# lambda.factor = 1e-20
# mu_ratio = 1
# outer_eps = 1e-6
# keep = TRUE
# Trange = c(0,1)
rule <- "lam.min"
# cv_cgl, Constrained group lasso
cv_cgl <- cv.FuncompCGL(y = y, X = X, Zc = Zc, intercept = intercept,
W = rep(1, p), #W = function(x){ diag( 1 / apply(x, 2, sd) ) },
k = k_list, trim = 0,
foldid = foldid, #nfolds = 10,
tol = 0, inner_eps = 1e-6, inner_maxiter = 1E3,
dfmax = 30, lambda.factor = 1e-3,
mu_ratio = 1, outer_eps = 1e-6,
keep = TRUE, Trange = c(0,1)
#lam = c(exp(seq(log(0.01),log(0.00001),length=100)),0)
)
plot(cv_cgl,k_list = k_list)
cv_cgl$Ftrim[c("lam.min", "lam.1se")]
beta <- coef(cv_cgl, trim = FALSE, s = rule)
k_opt <- (length(drop(beta)) - (m + 1)) / p
#cv_cgl$Funcomp.CGL.fit[[as.character(k_opt)]]
par(mfrow=c(1,4))
plot(cv_cgl,k_list = k_list)
plot(cv_cgl$Funcomp.CGL.fit[[as.character(k_opt)]],
p = p, k = k_opt, ylab = "L2")
plot(cv_cgl$Ftrim$cvm[k_opt - k_list[1] + 1, ])
title(paste0("k=", k_opt), line = 0.5)
# apply(cv_cgl$Funcomp.CGL.fit[[k_opt - 3]]$beta, 2,
# function(x, p, k) {
# #which(abs(x[seq(1, (p-1)*k+1, by = k)])>0),
# (1:p)[apply(matrix(x[1:(p*k)], byrow = TRUE, nrow = p), 1,
# function(x) max(abs(x)) > 0)]
# },p = p , k = k_opt)
if(k_opt == df_beta) {
plot(Data$beta, col = "red", pch = 19,
ylim = range(c(range(Data$beta), range(beta)))) #range(Data$beta))
abline(v= seq(from = 0, to = (p*df_beta), by = df_beta ))
abline(h = 0)
points(beta)
if(m1 > 0) points(p*df_beta + 1:m1, tail(Data$beta, m1),
col = "blue", pch = 19)
} else {
plot(beta, ylim = range(c(range(Data$beta), range(beta))) )
abline(v= seq(from = 0, to = (p*k_opt), by = k_opt ))
abline(h = 0, col = "red")
if(m1 > 0) points(p*k_opt + 1:m1, tail(Data$beta, m1),
col = "blue", pch = 19)
}
title(paste0("Method cgl"), outer=TRUE, line = -2)
par(mfrow=c(1,1))
beta_C <- matrix(beta[1:(p*k_opt)], byrow = TRUE, nrow = p)
cat("colSums:", colSums(beta_C))
#Non.zero <- which(abs(beta_C[,1]) > 0)
Non.zero <- apply(beta_C, 1, function(x) ifelse(max(abs(x)) >0, TRUE, FALSE))
Non.zero <- (1:p)[Non.zero]
cat("None zero groups:", Non.zero)
#vet(beta, p = p, k = k_opt)
par(mfrow=c(1,4))
plot(cv_cgl) #plot(cv_cgl,k_list = k_list)
matplot(sseq, beta_curve.true,
ylab = "coeffcients curve", xlab = "TIME", #main = "TRUE",
ylim = range(Data$beta[1:(p*df_beta)]),
type = "l")
abline(a = 0, b = 0, col = "grey", lwd = 2)
title("TRUE", line = 0.5)
text(0, beta_curve.true[1, Non_zero.true], labels = paste(Non_zero.true))
B <- splines::bs(Data$basis.info[,1], df = k_opt, intercept = TRUE)
beta_curve <- B %*% t(beta_C)
matplot(sseq, beta_curve,
ylab = "coef", xlab = "TIME", #main = "ESTI",
ylim = range(Data$beta[1:(p*df_beta)])#,
#type = "l"
)
abline(a = 0, b = 0, col = "grey", lwd = 2)
title("Estimate", line = 0.5)
text(0, beta_curve[1, Non.zero], labels = paste(Non.zero))
text(tail(sseq, 1), beta_curve[dim(beta_curve)[1], Non.zero], labels = paste(Non.zero))
plot(apply(abs(beta_C),1,sum))
title(paste0("k=", k_opt), line = 0.5)
title(paste0("Method cgl"), outer=TRUE, line = -2)
par(mfrow=c(1,1))
##set a cutoff when you compute nonzeros
Non.zero <- apply(beta_C, 1, function(x)
ifelse(sqrt(sum(x^2)) > sqrt(sum(beta_C^2))/100, TRUE, FALSE))
Non.zero <- (1:p)[Non.zero]
Non.zero
Non.zero <- apply(beta_C, 1, function(x)
ifelse(sum(x^2) > sum(beta_C^2)/100, TRUE, FALSE))
Non.zero <- (1:p)[Non.zero]
Non.zero
## cut by curve
Curve_L2 <- colSums(beta_curve^2)
Curve_L2 <- Curve_L2 - colSums(beta_curve[c(1, nrow(Data$basis.info)), ]^2) / 2
Curve_L2 <- Curve_L2 * (Data$basis.info[2,1] - Data$basis.info[1,1])
plot(Curve_L2)
which(sqrt(Curve_L2) > sqrt(sum(Curve_L2)) / 100)
cgl <- list()
# MSE <- crossprod(y - cbind2(cbind(cv_cgl$Funcomp.CGL.fit[[k_opt - k_list[1]+ 1]]$Z,
# Zc), 1) %*% beta) / length(y)
X_train <- cbind2(cbind(cv_cgl$Funcomp.CGL.fit[[k_opt - k_list[1]+ 1]]$Z, Zc), 1)
MSE <- sum((y - X_train %*% beta)^2) / length(y)
# R_sqr <- 1 - crossprod(y - cbind2(cbind(cv_cgl$Funcomp.CGL.fit[[k_opt - k_list[1]+ 1]]$Z,
# Zc), 1) %*% beta) / crossprod(y - mean(y))
R_sqr <- sum((y - X_train %*% beta)^2)
R_sqr <- 1 - R_sqr / crossprod(y - mean(y))
obj <- FuncompCGL(y = y_test, X = Test$data$Comp, k = k_opt, nlam = 1, outer_maxiter = 0)
# PE <- sum((Test$data$y - cbind2(cbind(obj$Z, Test$data$Zc), 1) %*% beta)^2)
# / length(drop(Test$data$y))
X_test <- cbind2(cbind(obj$Z, Test$data$Zc), 1)
PE <- sum((y_test - X_test %*% beta)^2) / length(y_test)
cgl$pred_error <- c(MSE = MSE, PE = PE, Rsqr_train = R_sqr)
cgl$Non.zero_cut <- Non.zero
cgl <- c(cgl,
ERROR_fun(beta_fit = beta, beta_true = Data$beta,
basis_fit = B, basis_true = Data$basis.info[,-1],
sseq = Data$basis.info[, 1],
m = m, p = p, Nzero_group = length(Non_zero.true), tol = 0),
k = k_opt)
cgl$coef <- list(beta_C = beta_C, beta_c = tail(beta, m1))
## Not run:
# naive model
# set mu_raio = 0 to identifying without linear constraints,
# no outer_loop for Lagrange augmented multiplier
# mu_ratio = 0
cv_naive <- cv.FuncompCGL(y = y, X = X, Zc = Zc, intercept = intercept,
W = rep(1, p), #W = function(x){ diag( 1 / apply(x, 2, sd) ) },
k = k_list, nfolds = 10, trim = 0,
tol = 0, inner_eps = 1e-6, inner_maxiter = 1E3,
dfmax = 30, lambda.factor = 1e-3,
mu_ratio = 0, outer_eps = 1e-6,
keep = FALSE, Trange = c(0,1)
#lam = c(exp(seq(log(0.01),log(0.00001),length=100)),0)
)
plot(cv_naive,k_list = k_list)
cv_naive$Ftrim[c("lam.min", "lam.1se")]
beta <- coef(cv_naive, trim = FALSE, s = rule)
k_opt <- (length(drop(beta)) - (m + 1)) / p
#cv_naive$Funcomp.CGL.fit[[as.character(k_opt)]]
par(mfrow=c(1,4))
plot(cv_naive,k_list = k_list)
plot(cv_naive$Funcomp.CGL.fit[[as.character(k_opt)]],
p = p, k = k_opt, ylab = "L2")
plot(cv_naive$Ftrim$cvm[k_opt - k_list[1] + 1, ])
title(paste0("k=", k_opt), line = 0.5)
# apply(cv_naive$Funcomp.CGL.fit[[k_opt - 3]]$beta, 2,
# function(x, p, k) {
# #which(abs(x[seq(1, (p-1)*k+1, by = k)])>0),
# (1:p)[apply(matrix(x[1:(p*k)], byrow = TRUE, nrow = p), 1,
# function(x) max(abs(x)) > 0)]
# },p = p , k = k_opt)
if(k_opt == df_beta) {
plot(Data$beta, col = "red", pch = 19,
ylim = range(c(range(Data$beta), range(beta)))) #range(Data$beta))
abline(v= seq(from = 0, to = (p*df_beta), by = df_beta ))
abline(h = 0)
points(beta)
if(m1 > 0) points(p*df_beta + 1:m1, tail(Data$beta, m1),
col = "blue", pch = 19)
} else {
plot(beta, ylim = range(c(range(Data$beta), range(beta))) )
abline(v= seq(from = 0, to = (p*k_opt), by = k_opt ))
abline(h = 0, col = "red")
if(m1 > 0) points(p*k_opt + 1:m1, tail(Data$beta, m1),
col = "blue", pch = 19)
}
title(paste0("Method naive"), outer=TRUE, line = -2)
par(mfrow=c(1,1))
beta_C <- matrix(beta[1:(p*k_opt)], byrow = TRUE, nrow = p)
cat("colSums:", colSums(beta_C))
#Non.zero <- which(abs(beta_C[,1]) > 0)
Non.zero <- apply(beta_C, 1, function(x) ifelse(max(abs(x)) >0, TRUE, FALSE))
Non.zero <- (1:p)[Non.zero]
cat("None zero groups:", Non.zero)
#vet(beta, p = p, k = k_opt)
par(mfrow=c(1,4))
plot(cv_naive) #plot(cv_naive,k_list = k_list)
matplot(sseq, beta_curve.true,
ylab = "coeffcients curve", xlab = "TIME", #main = "TRUE",
ylim = range(Data$beta[1:(p*df_beta)]),
type = "l")
abline(a = 0, b = 0, col = "grey", lwd = 2)
title("TRUE", line = 0.5)
text(0, beta_curve.true[1, Non_zero.true], labels = paste(Non_zero.true))
B <- splines::bs(Data$basis.info[,1], df = k_opt, intercept = TRUE)
beta_curve <- B %*% t(beta_C)
matplot(sseq, beta_curve,
ylab = "coef", xlab = "TIME", #main = "ESTI",
ylim = range(Data$beta[1:(p*df_beta)])#,
#type = "l"
)
abline(a = 0, b = 0, col = "grey", lwd = 2)
title("Estimate", line = 0.5)
text(0, beta_curve[1, Non.zero], labels = paste(Non.zero))
text(tail(sseq, 1), beta_curve[dim(beta_curve)[1], Non.zero], labels = paste(Non.zero))
plot(apply(abs(beta_C),1,sum))
title(paste0("k=", k_opt), line = 0.5)
title(paste0("Method naive"), outer=TRUE, line = -2)
par(mfrow=c(1,1))
##set a cutoff when you compute nonzeros
Non.zero <- apply(beta_C, 1, function(x)
ifelse(sqrt(sum(x^2)) > sqrt(sum(beta_C^2))/100, TRUE, FALSE))
Non.zero <- (1:p)[Non.zero]
Non.zero
naive <- list()
# MSE <- crossprod(y - cbind2(cbind(cv_naive$Funcomp.CGL.fit[[k_opt - k_list[1]+ 1]]$Z,
# Zc), 1) %*% beta)
X_train <- cbind2(cbind(cv_naive$Funcomp.CGL.fit[[k_opt - k_list[1]+ 1]]$Z, Zc), 1)
MSE <- sum((y - X_train %*% beta)^2)/ length(y)
# R_sqr <- 1 - crossprod(y - cbind2(cbind(cv_naive$Funcomp.CGL.fit[[k_opt - k_list[1]+ 1]]$Z,
# Zc), 1) %*% beta) / crossprod(y - mean(y))
R_sqr <- sum((y - X_train %*% beta)^2)
R_sqr <- 1 - R_sqr / crossprod(y - mean(y))
obj <- FuncompCGL(y = Test$data$y, X = Test$data$Comp, k = k_opt, nlam = 1, outer_maxiter = 0)
# PE <- sum((Test$data$y - cbind2(cbind(obj$Z, Test$data$Zc), 1) %*% beta)^2)
# / length(drop(Test$data$y))
X_test <- cbind2(cbind(obj$Z, Test$data$Zc), 1)
PE <- sum((y_test - X_test %*% beta)^2) / length(y_test)
naive$pred_error <- c(MSE = MSE, PE = PE, Rsqr_train = R_sqr)
naive$Non.zero_cut <- Non.zero
naive <- c(naive,
ERROR_fun(beta_fit = beta, beta_true = Data$beta,
basis_fit = B, basis_true = Data$basis.info[,-1],
sseq = Data$basis.info[, 1],
m = m, p = p, Nzero_group = length(Non_zero.true), tol = 0),
k = k_opt)
naive$coef <- list(beta_C = beta_C, beta_c = tail(beta, m1))
# log contract model
# set reference variable and once ref is set to a scalar in range,
# mu_ratio is set to 0 automatically
# ref = sample(1:p, 1)
cv_base <- cv.FuncompCGL(y = y, X = X, Zc = Zc, intercept = intercept, ref = sample(1:p, 1),
W = rep(1, p - 1), #W = function(x){ diag( 1 / apply(x, 2, sd) ) },
k = k_list, nfolds = 10, trim = 0,
tol = 0, inner_eps = 1e-6, inner_maxiter = 1E3,
dfmax = 30, lambda.factor = 1e-3,
mu_ratio = 0, outer_eps = 1e-6,
keep = FALSE, Trange = c(0,1)
#,lam = c(exp(seq(log(0.01),log(0.00001),length=100)),0)
)
plot(cv_base,k_list = k_list)
cv_base$Ftrim[c("lam.min", "lam.1se")]
beta <- coef(cv_base, trim = FALSE, s = rule)
k_opt <- (length(drop(beta)) - (m + 1)) / p
#cv_base$Funcomp.CGL.fit[[as.character(k_opt)]]
par(mfrow=c(1,4))
plot(cv_base,k_list = k_list)
plot(cv_base$Funcomp.CGL.fit[[as.character(k_opt)]],
p = p, k = k_opt, ylab = "L2")
plot(cv_base$Ftrim$cvm[k_opt - k_list[1] + 1, ])
title(paste0("k=", k_opt), line = 0.5)
# apply(cv_base$Funcomp.CGL.fit[[k_opt - 3]]$beta, 2,
# function(x, p, k) {
# #which(abs(x[seq(1, (p-1)*k+1, by = k)])>0),
# (1:p)[apply(matrix(x[1:(p*k)], byrow = TRUE, nrow = p), 1,
# function(x) max(abs(x)) > 0)]
# },p = p , k = k_opt)
if(k_opt == df_beta) {
plot(Data$beta, col = "red", pch = 19,
ylim = range(c(range(Data$beta), range(beta)))) #range(Data$beta))
abline(v= seq(from = 0, to = (p*df_beta), by = df_beta ))
abline(h = 0)
points(beta)
if(m1 > 0) points(p*df_beta + 1:m1, tail(Data$beta, m1),
col = "blue", pch = 19)
} else {
plot(beta, ylim = range(c(range(Data$beta), range(beta))) )
abline(v= seq(from = 0, to = (p*k_opt), by = k_opt ))
abline(h = 0, col = "red")
if(m1 > 0) points(p*k_opt + 1:m1, tail(Data$beta, m1),
col = "blue", pch = 19)
}
title(paste0("Method baseline=", cv_base$Funcomp.CGL.fit[[1]]$ref), outer=TRUE, line = -2)
par(mfrow=c(1,1))
beta_C <- matrix(beta[1:(p*k_opt)], byrow = TRUE, nrow = p)
cat("colSums:", colSums(beta_C))
#Non.zero <- which(abs(beta_C[,1]) > 0)
Non.zero <- apply(beta_C, 1, function(x) ifelse(max(abs(x)) >0, TRUE, FALSE))
Non.zero <- (1:p)[Non.zero]
cat("None zero groups:", Non.zero)
#vet(beta, p = p, k = k_opt)
par(mfrow=c(1,4))
plot(cv_base) #plot(cv_base,k_list = k_list)
matplot(sseq, beta_curve.true,
ylab = "coeffcients curve", xlab = "TIME", #main = "TRUE",
ylim = range(Data$beta[1:(p*df_beta)]),
type = "l")
abline(a = 0, b = 0, col = "grey", lwd = 2)
title("TRUE", line = 0.5)
text(0, beta_curve.true[1, Non_zero.true], labels = paste(Non_zero.true))
B <- splines::bs(Data$basis.info[,1], df = k_opt, intercept = TRUE)
beta_curve <- B %*% t(beta_C)
matplot(sseq, beta_curve,
ylab = "coef", xlab = "TIME", #main = "ESTI",
ylim = range(Data$beta[1:(p*df_beta)])#,
#type = "l"
)
abline(a = 0, b = 0, col = "grey", lwd = 2)
title("Estimate", line = 0.5)
text(0, beta_curve[1, Non.zero], labels = paste(Non.zero))
text(tail(sseq, 1), beta_curve[dim(beta_curve)[1], Non.zero], labels = paste(Non.zero))
plot(apply(abs(beta_C),1,sum))
title(paste0("k=", k_opt), line = 0.5)
title(paste0("Method baseline=", cv_base$Funcomp.CGL.fit[[1]]$ref), outer=TRUE, line = -2)
par(mfrow=c(1,1))
##set a cutoff when you compute nonzeros
Non.zero <- apply(beta_C, 1, function(x)
ifelse(sqrt(sum(x^2)) > sqrt(sum(beta_C^2))/100, TRUE, FALSE))
Non.zero <- (1:p)[Non.zero]
Non.zero
base <- list()
# MSE <- crossprod(y - cbind2(cbind(cv_cgl$Funcomp.CGL.fit[[k_opt - k_list[1]+ 1]]$Z,
# Zc), 1) %*% beta) / length(y)
X_train <- cbind2(cbind(cv_cgl$Funcomp.CGL.fit[[k_opt - k_list[1]+ 1]]$Z, Zc), 1)
MSE <- crossprod(y - X_train %*% beta) / length(y)
# R_sqr <- 1 - crossprod(y - cbind2(cbind(cv_cgl$Funcomp.CGL.fit[[k_opt - k_list[1]+ 1]]$Z,
# Zc), 1) %*% beta) / crossprod(y - mean(y))
R_sqr <- crossprod(y - X_train %*% beta)
R_sqr <- 1 - R_sqr / crossprod(y - mean(y))
obj <- FuncompCGL(y = Test$data$y, X = Test$data$Comp, k = k_opt, nlam = 1, outer_maxiter = 0)
# PE <- sum((Test$data$y - cbind2(cbind(obj$Z, Test$data$Zc), 1) %*% beta)^2)
# / length(drop(Test$data$y))
X_test <- cbind2(cbind(obj$Z, Test$data$Zc), 1)
PE <- sum((y_test - X_test %*% beta)^2) / length(y_test)
base$pred_error <- c(MSE = MSE, PE = PE, Rsqr_train = R_sqr)
base$Non.zero_cut <- Non.zero
base <- c(base,
ERROR_fun(beta_fit = beta, beta_true = Data$beta,
basis_fit = B, basis_true = Data$basis.info[,-1],
sseq = Data$basis.info[, 1],
m = m, p = p, Nzero_group = length(Non_zero.true), tol = 0),
k = k_opt)
base$coef <- list(beta_C = beta_C, beta_c = tail(beta, m1))
## End(Not run)
|
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.