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
|
# File .../lmrobPredict.R
# Part of the R package 'robustbase', http://www.R-project.org
# Based on predict.lm (cf. src/library/stats/R/lm.R)
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 2 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# A copy of the GNU General Public License is available at
# http://www.r-project.org/Licenses/
# Note that '# *rob' indicate adjustment for the robust case
predict.lmrob <-
function(object, newdata, se.fit = FALSE, scale = NULL, df = NULL, # *rob
interval = c("none", "confidence", "prediction"),
level = .95, type = c("response", "terms"),
terms = NULL, na.action = na.pass, pred.var = res.var/weights,
weights = 1, ...)
{
tt <- terms(object)
if(!inherits(object, "lmrob") && !inherits(object, "glmrob")) # *rob
warning("calling predict.lm(<fake-lmrob-object>) ...") # *rob
if(missing(newdata) || is.null(newdata)) {
mm <- X <- model.matrix.lm(object)
mmDone <- TRUE
offset <- object$offset
}
else {
Terms <- delete.response(tt)
m <- model.frame(Terms, newdata, na.action = na.action,
xlev = object$xlevels)
if(!is.null(cl <- attr(Terms, "dataClasses"))) .checkMFClasses(cl, m)
X <- model.matrix(Terms, m, contrasts.arg = object$contrasts)
offset <- rep.int(0, nrow(X))
if (!is.null(off.num <- attr(tt, "offset")))
for(i in off.num)
offset <- offset + eval(attr(tt, "variables")[[i+1]], newdata)
if (!is.null(object$call$offset))
offset <- offset + eval(object$call$offset, newdata)
mmDone <- FALSE
}
n <- length(object$residuals) # NROW(qr(object)$qr)
p <- object$rank
## *rob: FIXME? predict.lm() here does pivoting and rank/ column selection
if(is.null(p)) df <- Inf # *rob
if(any(is.na(beta <- object$coefficients))) # *rob
stop("there are NAs in the coefficients") # *rob
predictor <- drop(X %*% beta) # *rob
if (!is.null(offset))
predictor <- predictor + offset
interval <- match.arg(interval)
if (interval == "prediction") {
if (missing(newdata)) { # *rob: this and next if statement are combined
warning("Predictions on current data refer to _future_ responses\n")
if (missing(weights)) {
w <- stats:::weights.default(object) # *rob
if (!is.null(w)) {
weights <- w
warning("Assuming prediction variance inversely proportional to weights used for fitting\n")
}
}
}
if (!missing(newdata) && missing(weights) && !is.null(object$weights) && missing(pred.var))
warning("Assuming constant prediction variance even though model fit is weighted\n")
if (inherits(weights, "formula")){
if (length(weights) != 2L)
stop("'weights' as formula should be one-sided")
d <- if(missing(newdata) || is.null(newdata))
model.frame(object)
else
newdata
weights <- eval(weights[[2L]], d, environment(weights))
}
}## "prediction" interval
type <- match.arg(type)
if(se.fit || interval != "none") {# *rob: whole 'then' statement is different
df <- object$df.residual
res.var <- if (is.null(scale)) object$s^2 else scale^2
if(type != "terms"){
ip <- diag(X %*% object$cov %*% t(X))
} else ip <- rep.int(0, n)
}
if (type == "terms") { ## type == "terms" ------------
if(!mmDone){
mm <- model.matrix.lm(object) # *rob: call of model.matrix.lm
# instead of model.matrix
mmDone <- TRUE
}
aa <- attr(mm, "assign")
ll <- attr(tt, "term.labels")
hasintercept <- attr(tt, "intercept") > 0L
if (hasintercept) ll <- c("(Intercept)", ll)
aaa <- factor(aa, labels = ll)
asgn <- split(order(aa), aaa)
if (hasintercept) {
asgn$"(Intercept)" <- NULL
if(!mmDone){
mm <- model.matrix.lm(object) # *rob: call of model.matrix.lm
# instead of model.matrix
mmDone <- TRUE
}
avx <- colMeans(mm)
termsconst <- sum(avx * beta) # *rob: no column selection
}
nterms <- length(asgn)
if(nterms > 0) {
predictor <- matrix(ncol = nterms, nrow = NROW(X))
dimnames(predictor) <- list(rownames(X), names(asgn))
if (se.fit || interval != "none") {
ip <- predictor # *rob: just this assignment is needed
}
if(hasintercept)
X <- sweep(X, 2L, avx, check.margin=FALSE)
unpiv <- seq_len(NCOL(X))
## *rob: Assuming there are no columns which are
## completely aliased with earlier columns.
for (i in seq.int(1L, nterms, length.out = nterms)) {
iipiv <- asgn[[i]] # Columns of X, ith term
ii <- unpiv[iipiv] # Corresponding rows of Rinv
iipiv[ii == 0L] <- 0L
predictor[, i] <-
if(any(iipiv > 0L)) X[, iipiv, drop = FALSE] %*% beta[iipiv]
else 0
if (se.fit || interval != "none"){
ip[, i] <- if(any(iipiv > 0L)){# *rob: next steps modified
h.X <- X[, iipiv, drop = FALSE]
diag(h.X %*%object$cov[iipiv, iipiv] %*% t(h.X))
} else 0
}
}
if (!is.null(terms)) {
predictor <- predictor[, terms, drop = FALSE]
if (se.fit)
ip <- ip[, terms, drop = FALSE]
}
} else { # no terms
predictor <- ip <- matrix(0, n, 0L)
}
attr(predictor, 'constant') <- if (hasintercept) termsconst else 0
}
### Now construct elements of the list that will be returned
if(interval != "none") {
tfrac <- qt((1 - level)/2, df)
hwid <- tfrac * switch(interval,
confidence = sqrt(ip),
prediction = sqrt(ip+pred.var)
)
if(type != "terms") {
predictor <- cbind(predictor, predictor + hwid %o% c(1, -1))
colnames(predictor) <- c("fit", "lwr", "upr")
} else {
if (!is.null(terms)) hwid <- hwid[, terms, drop = FALSE]
lwr <- predictor + hwid
upr <- predictor - hwid
}
}
if(se.fit || interval != "none") {
se <- sqrt(ip)
if (type == "terms" && !is.null(terms)) se <- se[, terms, drop = FALSE]
}
if(missing(newdata) && !is.null(na.act <- object$na.action)) {
predictor <- napredict(na.act, predictor)
if(se.fit) se <- napredict(na.act, se)
}
if(type == "terms" && interval != "none") {
if(missing(newdata) && !is.null(na.act)) {
lwr <- napredict(na.act, lwr)
upr <- napredict(na.act, upr)
}
list(fit = predictor, se.fit = se, lwr = lwr, upr = upr,
df = df, residual.scale = sqrt(res.var))
} else if (se.fit)
list(fit = predictor, se.fit = se,
df = df, residual.scale = sqrt(res.var))
else predictor
}
|
