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
|
# these are "simple" tests for generateFilterValuesData and filterFeatures.
# we test the general code here, for exhaustive tests for all filters please look at file
# "test_featsel_filters.R"
test_that("filterFeatures", {
# winbuilder gets stuck here
skip_on_cran()
ns = getTaskFeatureNames(binaryclass.task)
f = filterFeatures(binaryclass.task,
method = "variance", select = "threshold",
threshold = -Inf)
f$env = NULL
binaryclass.task.cp = binaryclass.task
binaryclass.task.cp$env = NULL
expect_equal(f, binaryclass.task.cp)
feat.imp.old = suppressWarnings(generateFilterValuesData(binaryclass.task))
expect_data_frame(feat.imp.old$data,
types = c("character", "numeric", "factor"),
nrows = length(ns), ncols = 4,
col.names = "named")
expect_equal(ns, feat.imp.old$data$name)
feat.imp.new = suppressWarnings(generateFilterValuesData(binaryclass.task))
expect_data_frame(feat.imp.new$data,
types = c("character", "numeric", "factor"),
nrows = length(ns), ncols = 4,
col.names = "named")
expect_equal(names(feat.imp.new$data), c("name", "type", "filter", "value"))
expect_equal(ns, feat.imp.new$data$name)
feat.imp.old = suppressWarnings(generateFilterValuesData(binaryclass.task,
method = "variance"))
expect_data_frame(feat.imp.old$data,
types = c("character", "numeric", "factor"),
nrows = length(ns), ncols = 4,
col.names = "named")
expect_equal(ns, feat.imp.old$data$name)
f = filterFeatures(binaryclass.task, method = "variance", abs = 5L)
expect_true(setequal(getTaskFeatureNames(f),
head(setorder(feat.imp.old$dat, -value), 5L)$name))
# now check that we get the same result by operating on
# generateFilterValuesData
feat.imp.old = suppressWarnings(generateFilterValuesData(binaryclass.task,
method = "variance"))
ff = filterFeatures(binaryclass.task, fval = feat.imp.old, abs = 5L)
f$env = NULL
ff$env = NULL
expect_equal(f, ff)
feat.imp.new = generateFilterValuesData(binaryclass.task, method = "variance")
expect_data_frame(feat.imp.new$data,
types = c("character", "numeric", "factor"),
nrow = length(ns), ncols = 4, col.names = "named")
expect_equal(names(feat.imp.new$data), c("name", "type", "filter", "value"))
expect_equal(ns, feat.imp.new$data$name)
f = filterFeatures(binaryclass.task, method = "variance", abs = 5L)
expect_true(setequal(
getTaskFeatureNames(f),
head(setorder(feat.imp.old$dat, -value), 5L)$name))
# now check that we get the same result by operating on generateFilterValuesData
feat.imp.new = generateFilterValuesData(binaryclass.task, method = "variance")
ff = filterFeatures(binaryclass.task, fval = feat.imp.new, abs = 5L)
f$env = NULL
ff$env = NULL
expect_equal(f, ff)
f1 = filterFeatures(binaryclass.task,
abs = 1L, mandatory.feat = "V1")
f2 = subsetTask(binaryclass.task, features = "V1")
f1$env = NULL
f2$env = NULL
expect_equal(f1, f2)
f1 = filterFeatures(multiclass.task, abs = round(0.5 * ncol(multiclass.df)))
f2 = filterFeatures(multiclass.task, perc = 0.5)
f1$env = NULL
f2$env = NULL
expect_equal(f1, f2)
lrn1 = makeFilterWrapper("classif.rpart", fw.perc = 0.2)
m = train(lrn1, multiclass.task)
f = getFilteredFeatures(m)
expect_type(f, "character")
expect_equal(length(f), 1L)
})
test_that("args are passed down to filter methods", {
# winbuilder gets stuck here
skip_on_cran()
# we had an issue here, see #941
expect_error(generateFilterValuesData(regr.num.task,
method = c("mrmr", "univariate.model.score"),
nselect = 3, perf.learner = "regr.lm"), "Please pass extra arguments")
# check that we can pass down perf.learner to univariate.model.score, and get
# no error from mrmr call
f = generateFilterValuesData(regr.num.task,
method = c("mrmr", "univariate.model.score"),
nselect = 3,
more.args = list(univariate.model.score = list(perf.learner = "regr.lm")))
# create stupid dummy data and check that we can change the na.rm arg of
# filter "variance" in multiple ways
d = iris
d[1L, 1L] = NA_real_
task = makeClassifTask(data = d, target = "Species")
f1 = generateFilterValuesData(task, method = "variance", na.rm = FALSE)
f2 = generateFilterValuesData(task, method = "variance", na.rm = TRUE)
f3 = generateFilterValuesData(task,
method = "variance",
more.args = list(variance = list(na.rm = TRUE)))
f4 = generateFilterValuesData(task,
method = c("univariate.model.score", "variance"),
more.args = list(variance = list(na.rm = TRUE)))
expect_true(is.na(f1$data$value[1L]))
expect_false(is.na(f2$data$value[1L]))
expect_false(is.na(f3$data$value[1L]))
expect_false(is.na(f4$data$value[1L]))
})
test_that("errors for unsupported task and feature types", {
# winbuilder gets stuck here
skip_on_cran()
expect_error(generateFilterValuesData(multiclass.task,
method = c("mrmr", "variance", "linear.correlation")),
"Filter(s) 'mrmr', 'linear.correlation' not compatible with task of type 'classif'",
fixed = TRUE)
expect_error(generateFilterValuesData(regr.task,
method = c("mrmr", "carscore")),
"Filter(s) 'mrmr', 'carscore' not compatible with features of type 'factors', and 'factors' respectively",
fixed = TRUE)
expect_error(generateFilterValuesData(regr.task,
method = "carscore"),
"Filter(s) 'carscore' not compatible with features of type 'factors' respectively",
fixed = TRUE)
})
test_that("filter values are named and ordered correctly", {
# winbuilder gets stuck here
skip_on_cran()
# we had an issue here, see #940
ns = getTaskFeatureNames(regr.task)
mock.filter = makeFilter(
"mock.filter",
desc = "Mock Filter",
pkg = character(0),
supported.tasks = c("classif", "regr", "surv"),
supported.features = c("numerics", "factors"),
fun = function(task, nselect) {
ns = getTaskFeatureNames(task)
d = seq_along(ns)
names(d) = ns
d = c(d[-1], d[1])
d
}
)
fv = generateFilterValuesData(regr.task, method = "mock.filter")
expect_equal(fv$data$name, ns)
expect_equal(fv$data$value, seq_along(ns))
rm("mock.filter", envir = mlr:::.FilterRegister)
})
test_that("filter method 'variance' works with missing values", {
# winbuilder gets stuck here
skip_on_cran()
fi = generateFilterValuesData(regr.na.num.task, method = "variance")
expect_false(anyMissing(fi$data$value))
})
test_that("ensemble methods work", {
# winbuilder gets stuck here
skip_on_cran()
fi = generateFilterValuesData(multiclass.task,
method = list("E-min", c(
"FSelectorRcpp_gain.ratio",
"FSelectorRcpp_information.gain")))
expect_true(all(!is.na(fi$data$value) == TRUE))
})
|
