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
|
library(cluster)
## Compare on these:
nms <- c("clustering", "objective", "isolation", "clusinfo", "silinfo")
nm2 <- c("medoids", "id.med", nms)
nm3 <- nm2[- pmatch("obj", nm2)]
(x <- x0 <- cbind(V1 = (-3:4)^2, V2 = c(0:6,NA), V3 = c(1,2,NA,7,NA,8:9,8)))
(px <- pam(x,2, metric="manhattan"))
stopifnot(identical(x,x0))# DUP=FALSE ..
pd <- pam(dist(x,"manhattan"), 2)
px2 <- pam(x,2, metric="manhattan", keep.diss=FALSE, keep.data=FALSE)
pdC <- pam(x,2, metric="manhattan", cluster.only = TRUE)
p1 <- pam(x,1, metric="manhattan")
stopifnot(identical(px[nms], pd[nms]),
identical(px[nms], px2[nms]),
identical(pdC, px2$clustering),
## and for default dist "euclidean":
identical(pam(x, 2)[nms],
pam(dist(x),2)[nms]),
identical(p1[c("id.med", "objective", "clusinfo")],
list(id.med = 6L, objective = c(build=9.25, swap=9.25),
clusinfo = array(c(8, 18, 9.25, 45, 0), dim = c(1, 5),
dimnames=list(NULL, c("size", "max_diss", "av_diss",
"diameter", "separation"))))),
p1$clustering == 1, is.null(p1$silinfo)
)
set.seed(253)
## generate 250 objects, divided into 2 clusters.
x <- rbind(cbind(rnorm(120, 0,8), rnorm(120, 0,8)),
cbind(rnorm(130,50,8), rnorm(130,10,8)))
.proctime00 <- proc.time()
summary(px2 <- pam(x, 2))
pdx <- pam(dist(x), 2)
all.equal(px2[nms], pdx[nms], tol = 1e-12) ## TRUE
pdxK <- pam(dist(x), 2, keep.diss = TRUE)
stopifnot(identical(pdx[nm2], pdxK[nm2]))
spdx <- silhouette(pdx)
summary(spdx)
spdx
postscript("pam-tst.ps")
if(FALSE)
plot(spdx)# the silhouette
## is now identical :
plot(pdx)# failed in 1.7.0 -- now only does silhouette
par(mfrow = 2:1)
## new 'dist' argument for clusplot():
plot(pdx, dist=dist(x))
## but this should work automagically (via eval()) as well:
plot(pdx)
## or this
clusplot(pdx)
data(ruspini)
summary(pr4 <- pam(ruspini, 4))
(pr3 <- pam(ruspini, 3))
(pr5 <- pam(ruspini, 5))
data(votes.repub)
summary(pv3 <- pam(votes.repub, 3))
(pv4 <- pam(votes.repub, 4))
(pv6 <- pam(votes.repub, 6, trace = 3))
cat('Time elapsed: ', proc.time() - .proctime00,'\n')
## re-starting with medoids from pv6 shouldn't change:
pv6. <- pam(votes.repub, 6, medoids = pv6$id.med, trace = 3)
identical(pv6[nm3], pv6.[nm3])
## This example seg.faulted at some point:
d.st <- data.frame(V1= c(9, 12, 12, 15, 9, 9, 13, 11, 15, 10, 13, 13,
13, 15, 8, 13, 13, 10, 7, 9, 6, 11, 3),
V2= c(5, 9, 3, 5, 1, 1, 2, NA, 10, 1, 4, 7,
4, NA, NA, 5, 2, 4, 3, 3, 6, 1, 1),
V3 = c(63, 41, 59, 50, 290, 226, 60, 36, 32, 121, 70, 51,
79, 32, 42, 39, 76, 60, 56, 88, 57, 309, 254),
V4 = c(146, 43, 78, 88, 314, 149, 78, NA, 238, 153, 159, 222,
203, NA, NA, 74, 100, 111, 9, 180, 50, 256, 107))
dd <- daisy(d.st, stand = TRUE)
(r0 <- pam(dd, 5))# cluster 5 = { 23 } -- on single observation
## pam doing the "daisy" computation internally:
r0s <- pam(d.st, 5, stand=TRUE, keep.diss=FALSE, keep.data=FALSE)
(ii <- which(names(r0) %in% c("call","medoids")))
stopifnot(all.equal(r0[-ii], r0s[-ii], tol=1e-14),
identical(r0s$medoids, data.matrix(d.st)[r0$medoids, ]))
## This gave only 3 different medoids -> and seg.fault:
(r5 <- pam(dd, 5, medoids = c(1,3,20,2,5), trace = 2)) # now "fine"
dev.off()
##------------------------ Testing pam() with new "pamonce" argument:
## This is from "next version of Matrix" test-tools-1.R:
showSys.time <- function(expr) {
## prepend 'Time' for R CMD Rdiff
st <- system.time(expr)
writeLines(paste("Time", capture.output(print(st))))
invisible(st)
}
show6Ratios <- function(...) {
stopifnot(length(rgs <- list(...)) == 6,
nchar(ns <- names(rgs)) > 0)
r <- round(cbind(..1, ..2, ..3, ..4, ..5, ..6)[c(1,5),], 5)
dimnames(r) <- list(paste("Time ", rownames(r)), ns)
r
}
n <- 1000
## If not enough cases, all CPU times equals 0.
n <- 500 # for now, and automatic testing
sd <- 0.5
set.seed(13)
n2 <- as.integer(round(n * 1.5))
x <- rbind(cbind(rnorm( n,0,sd), rnorm( n,0,sd)),
cbind(rnorm(n2,5,sd), rnorm(n2,5,sd)),
cbind(rnorm(n2,7,sd), rnorm(n2,7,sd)),
cbind(rnorm(n2,9,sd), rnorm(n2,9,sd)))
## original algorithm
st0 <- showSys.time(pamx <- pam(x, 4, trace.lev=2))# 8.157 0.024 8.233
## bswapPamOnce algorithm
st1 <- showSys.time(pamxonce <- pam(x, 4, pamonce=TRUE, trace.lev=2))# 6.122 0.024 6.181
## bswapPamOnceDistIndice
st2 <- showSys.time(pamxonce2 <- pam(x, 4, pamonce = 2, trace.lev=2))# 4.101 0.024 4.151
## bswapPamSchubert FastPAM1
st3 <- showSys.time(pamxonce3 <- pam(x, 4, pamonce = 3, trace.lev=2))#
## bswapPamSchubert FastPAM2
st4 <- showSys.time(pamxonce4 <- pam(x, 4, pamonce = 4, trace.lev=2))#
## bswapPamSchubert FastPAM2 with linearized memory access
st5 <- showSys.time(pamxonce5 <- pam(x, 4, pamonce = 5, trace.lev=2))#
## bswapPamSchubert FasterPAM
st6 <- showSys.time(pamxonce6 <- pam(x, 4, pamonce = 6, trace.lev=2))#
show6Ratios('6:orig' = st6/st0, '5:orig' = st5/st0, '4:orig' = st4/st0, '3:orig' = st3/st0, '2:orig' = st2/st0, '1:orig' = st1/st0)
## only call element is not equal
(icall <- which(names(pamx) == "call"))
pamx[[icall]]
stopifnot(all.equal(pamx [-icall], pamxonce [-icall]),
all.equal(pamxonce[-icall], pamxonce2[-icall]),
all.equal(pamxonce[-icall], pamxonce3[-icall]),
all.equal(pamxonce[-icall], pamxonce4[-icall]),
all.equal(pamxonce[-icall], pamxonce5[-icall]),
all.equal(pamxonce[-icall], pamxonce6[-icall]))
## Same using specified medoids
(med0 <- 1 + round(n* c(0,1, 2.5, 4)))# lynne (~ 2010, AMD Phenom II X4 925)
st0 <- showSys.time(pamxst <- pam(x, 4, medoids = med0, trace.lev=2))# 13.071 0.024 13.177
st1 <- showSys.time(pamxoncest <- pam(x, 4, medoids = med0, pamonce=TRUE, trace.lev=2))# 8.503 0.024 8.578
st2 <- showSys.time(pamxonce2st <- pam(x, 4, medoids = med0, pamonce=2, trace.lev=2))# 5.587 0.025 5.647
st3 <- showSys.time(pamxonce3st <- pam(x, 4, medoids = med0, pamonce=3, trace.lev=2))#
st4 <- showSys.time(pamxonce4st <- pam(x, 4, medoids = med0, pamonce=4, trace.lev=2))#
st5 <- showSys.time(pamxonce5st <- pam(x, 4, medoids = med0, pamonce=5, trace.lev=2))#
st6 <- showSys.time(pamxonce6st <- pam(x, 4, medoids = med0, pamonce=6, trace.lev=2))#
show6Ratios('6:orig' = st6/st0, '5:orig' = st5/st0, '4:orig' = st4/st0, '3:orig' = st3/st0, '2:orig' = st2/st0, '1:orig' = st1/st0)
## only call element is not equal
stopifnot(all.equal(pamxst [-icall], pamxoncest [-icall]),
all.equal(pamxoncest[-icall], pamxonce2st[-icall]),
all.equal(pamxoncest[-icall], pamxonce3st[-icall]),
all.equal(pamxoncest[-icall], pamxonce4st[-icall]),
all.equal(pamxoncest[-icall], pamxonce5st[-icall]),
all.equal(pamxoncest[-icall], pamxonce6st[-icall]))
## Different starting values
med0 <- 1:4 # lynne (~ 2010, AMD Phenom II X4 925)
st0 <- showSys.time(pamxst <- pam(x, 4, medoids = med0, trace.lev=2))# 13.416 0.023 13.529
st1 <- showSys.time(pamxoncest <- pam(x, 4, medoids = med0, pamonce=TRUE, trace.lev=2))# 8.384 0.024 8.459
st2 <- showSys.time(pamxonce2st <- pam(x, 4, medoids = med0, pamonce=2, trace.lev=2))# 5.455 0.030 5.520
st3 <- showSys.time(pamxonce3st <- pam(x, 4, medoids = med0, pamonce=3, trace.lev=2))#
st4 <- showSys.time(pamxonce4st <- pam(x, 4, medoids = med0, pamonce=4, trace.lev=2))#
st5 <- showSys.time(pamxonce5st <- pam(x, 4, medoids = med0, pamonce=5, trace.lev=2))#
st6 <- showSys.time(pamxonce6st <- pam(x, 4, medoids = med0, pamonce=6, trace.lev=2))#
show6Ratios('6:orig' = st6/st0, '5:orig' = st5/st0, '4:orig' = st4/st0, '3:orig' = st3/st0, '2:orig' = st2/st0, '1:orig' = st1/st0)
## only call element is not equal
stopifnot(all.equal(pamxst [-icall], pamxoncest [-icall]),
all.equal(pamxoncest[-icall], pamxonce2st[-icall]),
all.equal(pamxoncest[-icall], pamxonce3st[-icall]),
all.equal(pamxoncest[-icall], pamxonce4st[-icall]),
all.equal(pamxoncest[-icall], pamxonce5st[-icall]),
all.equal(pamxoncest[-icall], pamxonce6st[-icall]))
## Medoid bug --- MM: Fixed, well "0L+ hack", in my pam.q, on 2012-01-31
## ----------
med0 <- (1:6)
st0 <- showSys.time(pamxst <- pam(x, 6, medoids = med0 , trace.lev=2))
stopifnot(identical(med0, 1:6))
med0 <- (1:6)
st1 <- showSys.time(pamxst.1 <- pam(x, 6, medoids = med0 , pamonce=1, trace.lev=2))
stopifnot(identical(med0, 1:6))
med0 <- (1:6)
st2 <- showSys.time(pamxst.2 <- pam(x, 6, medoids = med0 , pamonce=2, trace.lev=2))
stopifnot(identical(med0, 1:6))
med0 <- (1:6)
st3 <- showSys.time(pamxst.3 <- pam(x, 6, medoids = med0 , pamonce=3, trace.lev=2))
stopifnot(identical(med0, 1:6))
med0 <- (1:6)
st4 <- showSys.time(pamxst.4 <- pam(x, 6, medoids = med0 , pamonce=4, trace.lev=2))
stopifnot(identical(med0, 1:6))
med0 <- (1:6)
st5 <- showSys.time(pamxst.5 <- pam(x, 6, medoids = med0 , pamonce=5, trace.lev=2))
stopifnot(identical(med0, 1:6))
med0 <- (1:6)
st6 <- showSys.time(pamxst.6 <- pam(x, 6, medoids = med0 , pamonce=6, trace.lev=2))
stopifnot(identical(med0, 1:6))
stopifnot(all.equal(pamxst[-icall], pamxst.1 [-icall]),
all.equal(pamxst[-icall], pamxst.2 [-icall]),
all.equal(pamxst[-icall], pamxst.3 [-icall]),
all.equal(pamxst[-icall], pamxst.4 [-icall]),
all.equal(pamxst[-icall], pamxst.5 [-icall]))
# FasterPAM finds a better solution here, by chance
stopifnot(pamxst$objective >= pamxst.6$objective)
## Last Line:
cat('Time elapsed: ', proc.time() - .proctime00,'\n')
|
