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
|
library(lme4)
library(testthat)
(testLevel <- lme4:::testLevel())
L <- load(system.file("testdata/lme-tst-fits.rda",
package="lme4", mustWork=TRUE))
if (testLevel>1) {
if (getRversion() > "3.0.0") {
## saved fits are not safe with old R versions
fm1 <- fit_sleepstudy_1
s1 <- simulate(fm1,seed=101)[[1]]
s2 <- simulate(fm1,seed=101,use.u=TRUE)
s3 <- simulate(fm1,seed=101,nsim=10)
s4 <- simulate(fm1,seed=101,use.u=TRUE,nsim=10)
stopifnot(length(s3)==10,all(sapply(s3,length)==180),
length(s4)==10,all(sapply(s4,length)==180))
# test hook for cluster random effects
fakerand <- function(n) seq(from=1.0, by=0.4, length.out=n)
# In fact there are 10 observations/subject for each, but more robustly
ns <- tabulate(model.frame(fm1)$Subject)
s5 <- simulate(fm1, nsim=1, seed=12345, cluster.rand=fakerand)
su <- as.data.frame(VarCorr(fm1))[1, "sdcor"] # sd if cluster effects
y1 <- predict(fm1, re.form = ~0, se.fit=FALSE) # fixed effect
y2 <- su*rep(fakerand(length(ns)), times = ns) # cluster effect, per individual
set.seed(12345)
y3 <- sigma(fm1)*rnorm(length(y1)) # individual error terms
y <- y1+y2+y3
# testthat::expect_equal automatically incorporates the tolerance *if*
# it is in 3rd edition mode. Currently, this package does not use that mode.
stopifnot(all(abs(s5-y)< testthat_tolerance()))
## binomial (2-column and prob/weights)
gm1 <- fit_cbpp_1
gm2 <- fit_cbpp_3
gm1_s1 <- simulate(gm1,seed=101)[[1]]
gm1_s2 <- simulate(gm2,seed=101)[[1]]
stopifnot(all.equal(gm1_s1[,1]/rowSums(gm1_s1),gm1_s2))
gm1_s3 <- simulate(gm1,seed=101,use.u=TRUE)
gm1_s4 <- simulate(gm1,seed=101,nsim=10)
gm1_s5 <- simulate(gm2,seed=101,nsim=10)
stopifnot(length(gm1_s4)==10,all(sapply(gm1_s4,ncol)==2),all(sapply(gm1_s4,nrow)==56))
stopifnot(length(gm1_s5)==10,all(sapply(gm1_s5,length)==56))
## binomial (factor): Kubovy bug report 1 Aug 2013
d <- data.frame(y=factor(rep(letters[1:2],each=100)),
f=factor(rep(1:10,10)))
g1 <- glmer(y~(1|f),data=d,family=binomial)
s6 <- simulate(g1,nsim=10)
stopifnot(length(s6)==10,all(sapply(s6,length)==200))
## test explicitly stated link function
gm3 <- glmer(cbind(incidence, size - incidence) ~ period +
(1 | herd), data = cbpp, family = binomial(link="logit"))
s4 <- simulate(gm3,seed=101)[[1]]
stopifnot(all.equal(gm1_s1,s4))
cbpp$obs <- factor(seq(nrow(cbpp)))
gm4 <- fit_cbpp_2
## glmer(cbind(incidence, size - incidence) ~ period +
## (1 | herd) + (1|obs), data = cbpp, family = binomial)
s5 <- simulate(gm4,seed=101)[[1]]
s6 <- simulate(gm4,seed=101,use.u=TRUE)[[1]]
## Bernoulli
## works, but too slow
if (testLevel > 2) {
if(require("mlmRev")) {
data(guImmun, package="mlmRev")
table(guImmun$immun)
## N Y
## 1195 964
g1i <- glmer(immun ~ kid2p+mom25p+ord+ethn+momEd+husEd+momWork+rural+pcInd81+
(1|comm/mom), family="binomial", data=guImmun)
## In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv, :
## Model failed to converge with max|grad| = 0.326795 (tol = 0.002, component 1)
sg1 <- simulate(g1i)
if(FALSE) { ## similar: not relevant here {comment out for 'R CMD check'}:
## if(require("glmmTMB")) {
g2 <- glmmTMB(immun ~ kid2p+mom25p+ord+ethn+momEd+husEd+momWork+rural+pcInd81+
(1|comm/mom), family="binomial", data=guImmun)
sg2 <- simulate(g2)
## }
}
}
}
set.seed(101)
d <- data.frame(f = factor(rep(LETTERS[1:10],each=10)))
d$x <- runif(nrow(d))
u <- rnorm(10)
d$eta <- with(d, 1 + 2*x + u[f])
d$y <- rbinom(nrow(d), size=1, prob = plogis(d$eta))
g1 <- glmer(y ~ x + (1|f), data=d, family="binomial")
## tolPwrss=1e-5: no longer necessary
if (testLevel > 2) { ## trying a set of glmerControl(tolPwrss = 10^t) :
allcoef <- function(x) c(dev = deviance(x), th = getME(x,"theta"), beta = getME(x,"beta"))
tfun <- function(t) {
gg <- try( ## << errors (too small tolPwrss) are still printed :
glmer(y~x+(1|f),data=d,family="binomial",
control = glmerControl(tolPwrss = 10^t)))
if (inherits(gg,"try-error")) rep(NA,4) else allcoef(gg)
}
tvec <- seq(-4,-16,by=-0.25)
tres <- cbind(t = tvec, t(sapply(tvec, tfun)))
print(tres)
}
gm_s5 <- simulate(g1, seed=102)[[1]]
d$y <- factor(c("N","Y")[d$y+1])
g1B <- glmer(y ~ x + (1|f), data=d, family="binomial") ## ,tolPwrss=1e-5)
s1B <- simulate(g1B, seed=102)[[1]]
stopifnot(all.equal(gm_s5,as.numeric(s1B)-1))
## another Bernoulli
if(requireNamespace("mlmRev")) {
data(Contraception, package="mlmRev")
gm5 <- glmer(use ~ urban+age+livch+(1|district), Contraception, binomial)
s3 <- simulate(gm5)
}
d$y <- rpois(nrow(d),exp(d$eta))
gm6 <- glmer(y~x+(1|f),data=d,family="poisson")
s4 <- simulate(gm6)
## simulation 'from scratch' with formulas:
## binomial
## form <- formula(gm1)[-2]
form <- ~ (1|herd) + period
gm1_s4 <- simulate(form,newdata=model.frame(gm1),
newparams=list(theta=getME(gm1,"theta"),
beta=fixef(gm1)),
family=binomial,
weights=rowSums(model.frame(gm1)[[1]]),
seed=101)[[1]]
stopifnot(all.equal(gm1_s2,gm1_s4))
gm1_s5 <- simulate(formula(gm1),newdata=cbpp,
newparams=list(theta=getME(gm1,"theta"),
beta=fixef(gm1)),
family=binomial,
seed=101)[[1]]
stopifnot(all.equal(gm1_s1,gm1_s5))
tt <- getME(gm1,"theta")
bb <- fixef(gm1)
expect_error(simulate(form,newdata=model.frame(gm1),
newparams=list(theta=setNames(tt,"abc"),
beta=fixef(gm1)),
family=binomial,
weights=rowSums(model.frame(gm1)[[1]]),
seed=101),"mismatch between")
expect_error(simulate(form,newdata=model.frame(gm1),
newparams=list(theta=tt,
beta=setNames(bb,c("abc",names(bb)[-1]))),
family=binomial,
weights=rowSums(model.frame(gm1)[[1]]),
seed=101),"mismatch between")
## Gaussian
form <- formula(fm1)[-2]
s7 <- simulate(form,newdata=model.frame(fm1),
newparams=list(theta=getME(fm1,"theta"),
beta=fixef(fm1),
sigma=sigma(fm1)),
family=gaussian,
seed=101)[[1]]
stopifnot(all.equal(s7,s1))
## TO DO: wider range of tests, including offsets ...
}# R >= 3.0.0
} ## testLevel>1
|
