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
|
library(Matrix)
## This is example(sp....) -- much extended
mEQ <- function(x,y, ...) {
## first drop columns from y which are all 0 :
if(any(i0 <- colSums(abs(x)) == 0)) {
message(gettextf("x had %d zero-columns", sum(i0)))
x <- x[, !i0, drop=FALSE]
}
if(any(i0 <- colSums(abs(y)) == 0)) {
message(gettextf("y had %d zero-columns", sum(i0)))
y <- y[, !i0, drop=FALSE]
}
isTRUE(all.equal(x,y, tol=0, ...))
}
##' Is sparse.model.matrix() giving the "same" as dense model.matrix() ?
##'
##' @return logical
##' @param frml formula
##' @param dat data frame
##' @param showFactors
##' @param ...
isEQsparseDense <- function(frml, dat,
showFactors = isTRUE(getOption("verboseSparse")), ...)
{
## Author: Martin Maechler, Date: 21 Jul 2009
stopifnot(inherits(frml, "formula"), is.data.frame(dat))
if(showFactors)
print(attr(terms(frml, data=dat), "factors"))
smm <- sparse.model.matrix(frml, dat, ...)
mm <- model.matrix(frml, dat, ...)
sc <- smm@contrasts
mEQ(as(smm, "generalMatrix"), Matrix(mm, sparse=TRUE)) &
identical(smm@assign, attr(mm, "assign")) &
(if(is.null(mc <- attr(mm, "contrasts"))) length(sc) == 0 else identical(sc, mc))
}
### ------------ all the "datasets" we construct for use -------------
dd <- data.frame(a = gl(3,4), b = gl(4,1,12))# balanced 2-way
(dd3 <- cbind(dd, c = gl(2,6), d = gl(3,8)))
dd. <- dd3[- c(1, 13:15, 17), ]
set.seed(17)
dd4 <- cbind(dd, c = gl(2,6), d = gl(8,3))
dd4 <- cbind(dd4, x = round(rnorm(nrow(dd4)), 1))
dd4 <- dd4[- c(1, 13:15, 17), ]
##-> 'd' has unused levels
dM <- dd4
dM$X <- outer(10*rpois(nrow(dM), 2), 1:3)
dM$Y <- cbind(pmax(0, dM$x - .3), floor(4*rnorm(nrow(dM))))
str(dM)# contains *matrices*
options("contrasts") # the default: "contr.treatment"
op <- options(sparse.colnames = TRUE) # for convenience
stopifnot(identical(## non-sensical, but "should work" (with a warning each):
sparse.model.matrix(a~ 1, dd),
sparse.model.matrix( ~ 1, dd)))
sparse.model.matrix(~ a + b, dd, contrasts = list(a="contr.sum"))
sparse.model.matrix(~ a + b, dd, contrasts = list(b="contr.SAS"))
xm <- sparse.model.matrix(~ x, dM) # gives a warning, correctly
dxm <- Matrix(model.matrix(~ x, dM), sparse=TRUE)
stopifnot(is(xm, "sparseMatrix"), mEQ(as(xm,"generalMatrix"), dxm))
## Sparse method is equivalent to the traditional one :
stopifnot(isEQsparseDense(~ a + b, dd),
suppressWarnings(isEQsparseDense(~ x, dM)),
isEQsparseDense(~ 0 + a + b, dd),
identical(sparse.model.matrix(~ 0 + a + b, dd),
sparse.model.matrix(~ -1 + a + b, dd)),
isEQsparseDense(~ a + b, dd, contrasts = list(a="contr.sum")),
isEQsparseDense(~ a + b, dd, contrasts = list(a="contr.SAS")),
## contrasts as *functions* or contrast *matrices* :
isEQsparseDense(~ a + b, dd,
contrasts = list(a=contr.sum, b=contr.treatment(4))),
isEQsparseDense(~ a + b, dd, contrasts =
list(a=contr.SAS(3),# << ok after 'contrasts<-' update
b = function(n, contr=TRUE, sparse=FALSE)
contr.sum(n=n, contr=contr, sparse=sparse))))
sm <- sparse.model.matrix(~a * b, dd,
contrasts = list(a= contr.SAS(3, sparse = TRUE)))
sm
## FIXME: Move part of this to ../../MatrixModels/tests/
##stopifnot(all(sm == model.Matrix( ~a * b, dd, contrasts= list(a= contr.SAS(3)))))
##
stopifnot(isEQsparseDense(~ a + b + c + d, dd.))
stopifnot(isEQsparseDense(~ a + b:c + c + d, dd.))
## no intercept -- works too
stopifnot(isEQsparseDense(~ -1+ a + b + c + d, dd.))
stopifnot(isEQsparseDense(~ 0 + a + b:c + c + d, dd.))
Sparse.model.matrix <- function(...) {
s <- sparse.model.matrix(...)
as(s, "generalMatrix")# dropping 'assign',.. slots
}
##
dim(mm <- Matrix(model.matrix(~ a + b + c + d, dd4), sparse=TRUE))
dim(sm <- Sparse.model.matrix(~ a + b + c + d, dd4))
## dimension differ !!
stopifnot(mEQ(sm, mm)) ## but that's ok, since mm has all-0 column !
## look at this :
all(mm[,"d5"] == 0) ## !!!! --- correct: a column of all 0 <--> dropped level!
stopifnot(all.equal(sm, mm[, - which("d5" == colnames(mm))])) ## indeed !
## i.e., sm has just dropped an all zero column --- which it should!
stopifnot(isEQsparseDense(~ 1 + sin(x) + b*c + a:x, dd4, show=TRUE))
stopifnot(isEQsparseDense(~ I(a) + b*c + a:x, dd4, show=TRUE))
## no intercept -- works too
stopifnot(isEQsparseDense(~ 0+ I(a) + b*c + a:x, dd4, show=TRUE))
f <- ~ 1 + a + b*c + a*x
attr(terms(f, data=dd4), "factors")
dim(mm <- Matrix(model.matrix(f, data=dd4), sparse=TRUE))
dim(sm <- Sparse.model.matrix(f, data=dd4)) # ==
stopifnot(mEQ(sm, mm))
f <- ~ a*X + X*Y + a*c
attr(terms(f, data=dM), "factors")
dim(mm <- Matrix(model.matrix(f, data=dM), sparse=TRUE))
dim(sm <- Sparse.model.matrix(f, data=dM))
stopifnot(mEQ(sm, mm))
f <- ~ 1 + a + b*c + a*x + b*d*x + b:c:d
attr(terms(f, data=dd4), "factors")
dim(mm <- Matrix(model.matrix(f, data=dd4), sparse=TRUE)) ## 19 100
dim(sm <- Sparse.model.matrix(f, data=dd4)) # 19 88
stopifnot(mEQ(sm, mm))# {20 and 32 zero-columns ..}
## now get a bit courageous:
##
## stopifnot(isEQsparseDense(~ 1 + c + a:b:d, dat=dd4))
dim(mm <- Matrix(model.matrix(~ 1 + a + b*c + a:b:c:d, data=dd4),
sparse=TRUE)) ## 19 202
dim(sm <- Sparse.model.matrix(~ 1 + a + b*c + a:b:c:d, data=dd4)) # fails
stopifnot(mEQ(sm, mm))## {149 and 173 zero-columns !}
## stopifnot(isEQsparseDense(~ 1 + a + b*c + a:b:c:d, dat=dd4))
dim(mm <- Matrix(model.matrix(~ 1 + a + b:c + a:b:d, data=dd4),
sparse=TRUE)) ## 19 107
dim(sm <- Sparse.model.matrix(~ 1 + a + b:c + a:b:d, data=dd4)) # fails
stopifnot(mEQ(sm, mm))
dim(mm <- Matrix(model.matrix(~ a*b*c +c*d, dd4), sparse=TRUE)) ## 19 38
dim(sm <- Sparse.model.matrix(~ a*b*c +c*d, dd4))# 19 36
stopifnot(mEQ(sm, mm))
f1 <- ~ (a+b+c+d)^2 + (a+b):c:d + a:b:c:d
f2 <- ~ (a+b+c+d)^4 - a:b:c - a:b:d
mm1 <- Matrix(model.matrix(f1, dd4), sparse=TRUE)
dim(mm2 <- Matrix(model.matrix(f2, dd4), sparse=TRUE))
sm1 <- sparse.model.matrix(f1, dd4)
dim(sm2 <- sparse.model.matrix(f2, dd4))
stopifnot(identical(mm1,mm2),
identical(sm1,sm2),
mEQ(sm1, mm1))
str(dd <- data.frame(d = gl(10,6), a = ordered(gl(3,20))))
X. <- sparse.model.matrix(~ a + d, data = dd)
## failed because of contr.poly default in Matrix 0.999375-33
stopifnot(dim(X.) == c(60, 12), nnzero(X.) == 234,
isEQsparseDense(~ 0 + d + I(as.numeric(d)^2), dd))
## I(.) failed (upto 2010-05-07)
cat('Time elapsed: ', proc.time(),'\n') # for ``statistical reasons''
if(!interactive()) warnings()
|
