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
|
#### Utility functions for testing covMcd()
#### -------------------------------------- ../tests/tmcd.R
## "workhorse" -- by default *passed* to and called from doMCDdata():
domcd1 <- function(x, xname, nrep = 1,
## These are all got from doMCDdata() [yuck!]
method = get("method", parent.frame()), # compromise
time = get("time", parent.frame()), # compromise
short = get("short", parent.frame()), # compromise
full = get("full", parent.frame()), # compromise
lname = 20)
{
if(short && full)
stop("you should not set both 'full' and 'short' to TRUE")
force(xname)# => evaluate when it is a data(<>, ..) call
n <- dim(x)[1]
p <- dim(x)[2]
if(method == "MASS") {
mcd <- MASS::cov.mcd(x)
mcd$quan <- (n + p + 1) %/% 2 #default: floor((n+p+1)/2)
}
else if(method == "DetMCD") {
mcd <- covMcd(x, nsamp="deterministic") # trace = FALSE
}
else {
mcd <- covMcd(x) # trace = FALSE
}
if(full) {
header <- get("header", parent.frame())
header(time)
}
xres <- sprintf("%*s %3d %3d %3d %12.6f", lname, xname,
n, p, mcd$quan, mcd$crit)
if(time) {
xtime <- system.time(repMCD(x, nrep, method))[1]/nrep
xres <- sprintf("%s %10.1f", xres, 1000 * xtime)
}
cat(xres, "\n")
if(!short) {
cat("Best subsample: \n")
print(mcd$best)
ibad <- which(unname(mcd$mcd.wt) == 0)
nbad <- length(ibad)
cat("Outliers: ",nbad, if(nbad > 0)":", "\n")
if(nbad > 0)
print(ibad)
if(full) {
cat("------------- *MCD() result: --------------------------\n")
print(mcd)
}
cat("--------------------------------------------------------\n")
}
}## {domcd1}
##' Test the function covMcd() on the literature datasets:
##'
##' Call covMcd() for "all" datasets in robustbase / rrcov and print:
##' - execution time (if time is true)
##' - objective function
##' - best subsample found (if short is false)
##' - outliers identified (with cutoff 0.975) (if short is false)
##' - estimated center and covarinance matrix if full is true)
##'
##' @param nrep : [integer] number of repetitions to use for estimating the
##' (average) execution time
##' @param method : [character] select a method: one of (FASTMCD, MASS)
##' @param time : [logical] whether to evaluate the execution time
##' @param short : [logical] whether to do short output (i.e. only the
##' objective function value). If short == FALSE,
##' the best subsample and the identified outliers are
##' printed. See also the parameter full below
##' @param full : [logical] whether to print the estimated cente and covariance matrix
##' @param digits
##' @param domcd workhorse function, to be called e.g. as
##' @examples domcd(starsCYG, data(starsCYG), nrep)
##' @author Valentin Todorov; tweaks by Martin Maechler
##' @note Is called from ../../demo/determinMCD.R and ../../tests/tmcd.R
doMCDdata <- function(nrep = 1, method = c("FASTMCD", "MASS", "DetMCD"),
time = nrep >= 3, short = time, full = !short,
digits=5, domcd = domcd1)
{
stopifnot(is.function(domcd), length(formals(domcd)) >= 3)
options(digits = digits)
method <- match.arg(method) # *is* then accessed from domcd(.)
stopifnot(require("robustbase")) # all data() which do not specify package
data(Animals, package = "MASS")
brain <- Animals[c(1:24, 26:25, 27:28),]
data(list = c("fish", "pottery", "rice", "un86", "wages"), package = "rrcov")
tmp <- sys.call()
cat("\nCall: ", deparse(substitute(tmp)),"\n")
header <- function(time) { ## the string length here require 'lname <- 20' {FIXME}
## 1 2
## 1 3 5 7 901 3 5 7 90 2 4
cat("Data Set n p h(alf) LOG(obj)",if(time)" Time [ms]","\n",
"=============================================",if(time)"===========","\n",
sep="")
}
if(full) {
## header() is called in each domcd()
} else ## here
header(time)
domcd(bushfire, data(bushfire), nrep)
domcd(heart[, 1:2], data(heart), nrep)
domcd(starsCYG, data(starsCYG), nrep)
domcd(stack.x, data(stackloss), nrep)
domcd(data.matrix(subset(phosphor, select= -plant)),data(phosphor), nrep)
domcd(data.matrix(subset(coleman, select = -Y)), data(coleman), nrep)
domcd(data.matrix(subset(salinity, select = -Y)), data(salinity), nrep)
domcd(data.matrix(subset(wood, select = -y)), data(wood), nrep)
domcd(data.matrix(subset(hbk, select = -Y)), data(hbk), nrep)
domcd(brain, "Animals", nrep)
domcd(milk, data(milk), nrep)
domcd(lactic, data(lactic), nrep)
domcd(pension, data(pension), nrep)
domcd(pilot, data(pilot), nrep)
## This is for CovMcdBig ....
## domcd(radarImage, data(radarImage), nrep)
## domcd(NOxEmissions, data(NOxEmissions), nrep)
domcd(data.matrix(subset(vaso, select = -Y)), data(vaso), nrep)
domcd(data.matrix(subset(wagnerGrowth, select = -Period)), data(wagnerGrowth), nrep)
## Obs 14 has missing, column 7 is categorical
domcd(fish[-14,-7], data(fish, package="rrcov"), nrep)
domcd(pottery[,-7], data(pottery, package="rrcov"), nrep)
domcd(rice, data(rice, package="rrcov"), nrep)
domcd(un86, data(un86, package="rrcov"), nrep)
## there are missing values
domcd(wages[-c(29, 31, 38),-9], data(wages, package="rrcov"), nrep)
cat("========================================================\n")
} ## {doMCDdata}
if(FALSE){
data(mortality, package = "riv")
mm <- as.data.frame(lapply(mortality, signif, 3))
for(j in c(1,2,6,7))
mm[,j] <- mm[,j] * 10
mm[,5] <- mm[,5] * 1000
mm[,8] <- mm[,8] / 100
mort3 <- mm
dput(mort3)
}
## which gives the equivalent of
mort3 <-
data.frame(MO70 = c(140, 101, 86, 102, 115, 121, 118, 76.6,
131, 112, 111, 112, 117, 118, 123, 122, 81.7, 108, 111, 109,
92.5, 83.9, 93.8, 135, 124, 126, 122, 120, 127, 115, 156, 95.1,
127, 129, 116, 82.3, 115, 106, 134, 94.9, 119, 111, 131, 85.6,
135, 126, 141, 152, 137, 151, 93.6, 84.2, 78, 50.2, 81.3, 112,
80.1, 125, 120, 143),
MAGE = c(297, 277, 275, 268, 296, 327, 314, 258, 342, 278, 278,
313, 284, 272, 296, 277, 271, 296, 286, 250, 280, 270, 246, 301,
279, 287, 293, 271, 291, 295, 314, 267, 275, 307, 259, 251, 324,
285, 288, 254, 278, 287, 316, 287, 326, 309, 334, 369, 321, 311,
261, 272, 260, 244, 248, 277, 240, 295, 319, 346),
CI68 = c(137, 137, 129, 129, 151, 157, 157, 157, 157, 202, 202, 202,
138, 160, 190, 191, 191, 191, 159, 159, 146, 146, 203, 203, 182, 166,
203, 203, 167, 167, 165, 153, 149, 149, 149, 157, 152, 183, 183, 183,
183, 183, 183, 111, 171, 148, 148, 148, 192, 160, 160, 172, 172,
172, 172, 101, 173, 173, 144, 181),
MDOC = c(142, 80.4, 148, 167, 230, 187, 240, 149, 240, 195, 327,
377, 203, 160, 161, 68.7, 141, 120, 176, 105, 128, 112, 98.9, 160,
209, 200, 153, 126, 157, 157, 145, 160, 158, 102, 195, 188, 250,
143, 157, 186, 114, 129, 129, 143, 186, 207, 144, 112, 157, 121,
168, 155, 144, 144, 120, 194, 93.6, 231, 185, 89.7),
DENS = c(37, 37, 27, 32, 17, 13, 23, 19, 27, 29, 15, 15, 48, 34,
26, 47, 17, 10, 10, 18, 11, 13, 26, 19, 55, 17, 16, 7, 10, 17,
44, 13, 18, 26, 40, 22, 29, 7, 28, 10, 15, 1, 11, 10, 8, 13, 13,
6, 10, 26, 49, 28, 32, 18, 62, 15, 21, 18, 10, 12),
NONW = c(4.22, 3.36, 0.67, 0.52, 2.51, 0.82, 4.07, 1.11,
2.86, 2.92, 2.74, 1.05, 7.23, 5.16, 3.44, 2.84, 1.84,
1.47, 0.62, 0.03, 0.96, 1.07, 1.74, 2.41, 0.45, 4.7, 4.45,
1.2, 0.64, 2.28, 4.13, 1.06, 4.02, 2.22, 5.6, 0.43, 2.34,
1.78, 2.81, 1.9, 3.09, 1.43, 2.58, 1.34, 0.78, 3.44, 2.07,
0.68, 1, 3.6, 3.92, 2.58, 2.66, 0.05, 0.86, 0.32, 3.02,
4.24, 1.26, 1.08),
EDUC = c(454, 516, 601, 631, 565, 620, 661, 653, 661, 591,
568, 499, 685, 534, 539, 536, 560, 542, 680, 546, 648,
632, 601, 469, 458, 446, 521, 540, 661, 601, 480, 627,
506, 363, 551, 662, 518, 556, 484, 607, 562, 517, 521,
582, 629, 506, 534, 433, 459, 476, 492, 548, 517, 517,
468, 685, 483, 471, 678, 528),
IN69 = c(86.9, 99.3, 113, 99.2, 104, 118, 113, 117, 125,
100, 104, 115, 122, 107, 135, 101, 123, 114, 114, 113,
108, 109, 100, 99.8, 102, 100, 110, 112, 111, 113, 92.7,
116, 86.3, 103, 86.4, 109, 116, 112, 104, 108, 103, 116,
99.3, 116, 114, 104, 105, 97, 102, 83.4, 101, 125, 117,
118, 90.3, 108, 92.4, 106, 126, 109))
###'------*Generate* data for benchmarking ----------------------------------------
##' Generates a location contaminated multivariate
##' normal sample of n observations in p dimensions
##' (1-eps) * N_p(0, I_p) + eps * N_(m,I_p)
##' where
##' m = (b,b,...,b)
##' Defaults: eps=0 and b=10
##' @title Generate n x p location contaminated MV data
##' @param n number of observations
##' @param p number of variables
##' @param eps amount of contamination
##' @param b mean of "outliers"
gendata <- function(n,p, eps=0, b=10) {
if(missing(n) || missing(p))
stop("Please specify (n,p)")
if(!is.numeric(eps) || length(eps) != 1 || eps < 0 || eps >= 0.5)
stop("eps must be in [0,0.5)")
X <- matrix(rnorm(n*p), n, p)
nbad <- as.integer(eps * n)
if(nbad > 0) {
b <- rep(b, length = p) # recycle to p-vector
## = E[.] of bad obs.
xind <- sample(n,nbad)
X[xind,] <- X[xind, , drop=FALSE] + rep(b, each=nbad)
}
list(X=X, xind=if(nbad > 0) xind)
}
##' Repeated calls to different MCD algorithms for timing purposes *only*
repMCD <- function(x, nrep = 1, method = "FASTMCD") {
stopifnot(length(nrep) == 1, nrep >= 1)
switch(method,
"FASTMCD" = replicate(nrep, covMcd(x)),
"bestMCD" = replicate(nrep, covMcd(x, nsamp= "best")),
"exactMCD" = replicate(nrep, covMcd(x, nsamp= "exact")),
"DetMCD" = replicate(nrep, covMcd(x, nsamp="deterministic")),
"MASS.best" = replicate(nrep, MASS::cov.mcd(x)),# uses nsamp = "best" ==> up to 5000
## rrcov.control()$nsamp == 500 :
"MASS.500" = replicate(nrep, MASS::cov.mcd(x, nsamp = 500)),
## otherwise:
stop(gettextf("Method '%s' not yet implemented", method)))
}
repMCD.meths <- function() {
switch.expr <- body(repMCD)[[3]]
m <- names(switch.expr)
m[m != ""]
}
if(FALSE)
repMCD.meths()
## [1] "FASTMCD" "bestMCD" "DetMCD" "MASS.best" "MASS.500"
##' calls gendata(), repMCD()
dogen <- function(nrep=1, eps=0.49, method = repMCD.meths(), ## "FASTMCD" is first
p.set = c(2, 5, 10, 20, 30),
n.set = c(100, 500, 1000, 10000, 50000),
n.p.ratio = 5,
seed = 1234)
{
domcd <- function(x, nrep=1){
## system.time() *does* gc()
xtime <- system.time(repMCD(x, nrep, method))[1]/nrep
cat(sprintf("%6d %3d %12.2f\n", dim(x)[1], dim(x)[2], xtime))
xtime
}
set.seed(seed)
method <- match.arg(method)
mkL <- function(ch,m) paste(ch,m,sep="=")
ans <- matrix(NA, length(n.set), length(p.set),
dimnames = list(mkL("n",n.set), mkL("p",p.set)))
cat(sprintf("Method: %-12s; nrep = %d\n", method, nrep),
"------------------------------\n",
" n p Time\n",
"=======================\n", sep="")
for(n in n.set) {
n. <- mkL("n",n)
for(p in p.set) {
if(n.p.ratio * p <= n) {
xx <- gendata(n, p, eps)
ans[n., mkL("p",p)] <- domcd(xx$X, nrep)
}
}
}
cat("=======================\n")
cat(sprintf("Total time: %11.2f\n", nrep * sum(ans, na.rm=TRUE)))
structure(ans, nrep = nrep, method=method)
}## {dogen}
###' ------------------ These can only be used with rrcov :: CovMcd() --------------
docheck <- function(n, p, eps, ...) {
xx <- gendata(n,p,eps)
mcd <- CovMcd(xx$X, ...)
check(mcd, xx$xind)
}
##' check if mcd is robust w.r.t xind, i.e. check how many of xind
##' did not get zero weight
check <- function(mcd, xind){
mymatch <- xind %in% which(mcd@wt == 0)
length(xind) - sum(mymatch)
}
|
