-
Notifications
You must be signed in to change notification settings - Fork 0
/
01.data.extract.R
191 lines (142 loc) · 5.36 KB
/
01.data.extract.R
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
library("XML") # xmlSApply xmlRoot xmlParse xmlAttrs
library("bnlearn") # discretize
library("R.matlab") # readMat
library("RWeka") # write.arff
# ARFF to RDA
for (db in setdiff(conf.dbs, paste(conf.dbs.x2, "2", sep=""))) {
set.seed(conf.seed)
arff.file = sprintf("data/arff/%s/%s.arff", db, db)
xml.file = sprintf("data/arff/%s/%s.xml", db, db)
out.dir = sprintf("data/rda/%s", db)
dir.create(out.dir, showWarnings = FALSE, recursive = TRUE)
out.labels.file = sprintf("%s/%s.labels.rda", out.dir, db)
out.data.file = sprintf("%s/%s.data.rda", out.dir, db)
out.data.disc.file = sprintf("%s/%s.data.disc.rda", out.dir, db)
out.data.cont.file = sprintf("%s/%s.data.cont.rda", out.dir, db)
all.done = TRUE
if (!file.exists(out.labels.file) ||
!file.exists(out.data.file) ||
!file.exists(out.data.disc.file) ||
!file.exists(out.data.cont.file)) {
all.done = FALSE
}
for (i in 1:conf.nb.cv.reps) {
out.cv.file = sprintf("%s/%s.cv.splits.%02i.r%02i.rda", out.dir, db, conf.nb.cv, i)
if(!file.exists(out.cv.file)) {
all.done = FALSE
}
}
if (all.done) {
write.log(sprintf("%s - skipped", db))
next
}
t = proc.time()
# Image must be converted from matlab to arff format first
if (db == "image") {
m = readMat("data/arff/image/miml data.mat", verbose=FALSE, fixNames=TRUE)
# features
x = t(vapply(m$bags, function(x){as.vector(x[[1]])}, double(135)))
x = as.data.frame(x)
# labels
y = t(m$targets)
y[y==-1] = 0 # -1/1 -> 0/1
y = matrix(as.character(y), ncol = ncol(y))
y = as.data.frame(y)
# label names
targets = vapply(m$class.name, function(x){c(x[[1]])}, character(1))
df = cbind(x, y)
colnames(df) = c(paste("X", 1:135, sep=""), targets)
rownames(df) = NULL
write.arff(df, "data/arff/image/image.arff")
}
# recover label names
labels = as.character(xmlSApply(xmlRoot(xmlParse(xml.file)), function(x){xmlAttrs(x)["name"]}))
save(labels, file=out.labels.file)
data = read.arff(arff.file)
# Genbase "protein" column is a unique identifier (662 != values for 662 obs)
if (db == "genbase") {
data = data[, -1]
}
# Slashdot values are binary factors but encoded as numeric
if (db == "slashdot") {
data = data.frame(lapply(data, factor, levels=c(0, 1)))
}
save(data, file=out.data.file)
# transform everything to factor
data.disc = discretize(data, method="quantile", breaks=2)
save(data.disc, file=out.data.disc.file)
# transform everything (but labels) to numeric
data.cont = data
for (col in setdiff(colnames(data.cont), labels)) {
if (!is.numeric(data.cont[, col])) {
data.cont[, col] = as.numeric(data.cont[, col])
}
}
save(data.cont, file=out.data.cont.file)
for (i in 1:conf.nb.cv.reps) {
out.cv.file = sprintf("%s/%s.cv.splits.%02i.r%02i.rda", out.dir, db, conf.nb.cv, i)
cv.splits = sample(rep(1:conf.nb.cv, length.out = nrow(data)))
labels.ord = sample(length(labels))
save(cv.splits, labels.ord, file = out.cv.file)
}
t = proc.time() - t
write.log(sprintf("%s - done", db), t)
}
# Twin data sets
for (db in conf.dbs.x2) {
set.seed(conf.seed)
db2 = sprintf("%s2", db)
in.dir = sprintf("data/rda/%s", db)
in.labels.file = sprintf("%s/%s.labels.rda", in.dir, db)
in.data.file = sprintf("%s/%s.data.rda", in.dir, db)
in.data.disc.file = sprintf("%s/%s.data.disc.rda", in.dir, db)
in.data.cont.file = sprintf("%s/%s.data.cont.rda", in.dir, db)
out.dir = sprintf("data/rda/%s", db2)
dir.create(out.dir, showWarnings = FALSE, recursive = TRUE)
out.labels.file = sprintf("%s/%s.labels.rda", out.dir, db2)
out.data.file = sprintf("%s/%s.data.rda", out.dir, db2)
out.data.disc.file = sprintf("%s/%s.data.disc.rda", out.dir, db2)
out.data.cont.file = sprintf("%s/%s.data.cont.rda", out.dir, db2)
all.done = TRUE
if (!file.exists(out.labels.file) ||
!file.exists(out.data.file) ||
!file.exists(out.data.disc.file) ||
!file.exists(out.data.cont.file)) {
all.done = FALSE
}
for (i in 1:conf.nb.cv.reps) {
out.cv.file = sprintf("%s/%s.cv.splits.%02i.r%02i.rda", out.dir, db2, conf.nb.cv, i)
if(!file.exists(out.cv.file)) {
all.done = FALSE
}
}
if (all.done) {
write.log(sprintf("%s - skipped", db2))
next
}
t = proc.time()
load(in.labels.file)
load(in.data.file)
load(in.data.disc.file)
load(in.data.cont.file)
labels = colnames(data[, c(labels, labels)])
save(labels, file=out.labels.file)
s = sample(nrow(data))
data = cbind(data, data[s, ])
colnames(data) = colnames(data[, colnames(data)]) # trick
save(data, file=out.data.file)
data.disc = cbind(data.disc, data.disc[s, ])
colnames(data.disc) = colnames(data.disc[, colnames(data.disc)])
save(data.disc, file=out.data.disc.file)
data.cont = cbind(data.cont, data.cont[s, ])
colnames(data.cont) = colnames(data.cont[, colnames(data.cont)])
save(data.cont, file=out.data.cont.file)
for (i in 1:conf.nb.cv.reps) {
out.cv.file = sprintf("%s/%s.cv.splits.%02i.r%02i.rda", out.dir, db2, conf.nb.cv, i)
cv.splits = sample(rep(1:conf.nb.cv, length.out = nrow(data)))
labels.ord = sample(length(labels))
save(cv.splits, labels.ord, file = out.cv.file)
}
t = proc.time() - t
write.log(sprintf("%s - done", db2), t)
}