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
|
test_that("multilabel task", {
mt = multilabel.task
expect_equal(getTaskTargetNames(mt), c("y1", "y2"))
expect_equal(getTaskClassLevels(mt), c("y1", "y2"))
expect_equal(getTaskFormula(mt), y1 + y2 ~ .)
y = getTaskTargets(mt)
expect_true(is.data.frame(y) && ncol(y) == 2L)
expect_true(is.logical(y[, 1]) && is.logical(y[, 2L]))
expect_equal(colnames(y), c("y1", "y2"))
})
test_that("multilabel learning", {
lrn = makeLearner("multilabel.rFerns")
# train predict eval
mod = train(lrn, multilabel.task)
pred = predict(mod, multilabel.task)
p = performance(pred)
expect_true(!is.na(p))
pmulti = getMultilabelBinaryPerformances(pred, list(mmce, acc))
expect_true(is.matrix(pmulti))
expect_true(!any(is.na(pmulti)))
expect_equal(rownames(pmulti), getTaskTargetNames(multilabel.task))
expect_equal(colnames(pmulti), vcapply(list(mmce, acc), mlr:::measureAggrName))
# with newdata df
pred = predict(mod, newdata = multilabel.df)
p = performance(pred)
expect_true(!is.na(p))
pmulti = getMultilabelBinaryPerformances(pred, list(mmce, acc))
expect_true(!any(is.na(pmulti)))
# resample
r = holdout(lrn, multilabel.task)
expect_true(!is.na(r$aggr))
pmulti = getMultilabelBinaryPerformances(r$pred, list(mmce, acc))
expect_true(!any(is.na(pmulti)))
# Learner with Impute-Preprocessing
lrn = makeImputeWrapper(lrn, classes = list(integer = imputeMedian(), numeric = imputeMedian(), factor = imputeConstant("Const")))
multilabel.df2 = multilabel.df
multilabel.df2[c(2, 10, 14), c(1, 5)] = NA
multilabel.task2 = makeMultilabelTask("multilabel", data = multilabel.df2, target = multilabel.target)
mod = train(lrn, multilabel.task2)
pred = predict(mod, multilabel.task2)
p = performance(pred)
expect_true(!is.na(p))
# Learner with Hyperparameters
lrn = makeLearner("multilabel.rFerns", par.vals = list(depth = 6, ferns = 100))
mod = train(lrn, multilabel.task)
pred = predict(mod, multilabel.task)
p = performance(pred)
expect_true(!is.na(p))
})
test_that("MultilabelBinaryRelevanceWrapper with glmnet (#958)", {
requirePackagesOrSkip("glmnet", default.method = "load")
# multilabelBinaryRelevanceWrapper was not working properly for classif.glmnet, we had a bug here
lrn = makeLearner("classif.glmnet", predict.type = "response")
lrn2 = makeMultilabelBinaryRelevanceWrapper(lrn)
mod = train(lrn2, multilabel.task)
pred = predict(mod, multilabel.task)
expect_error(pred, NA)
})
testMultilabelWrapper = function(fun, ...) {
desc = fun("classif.rpart")$model.subclass[1]
test_that(desc, {
lrn1 = makeLearner("classif.rpart")
lrn2 = fun(lrn1, ...)
lrn2 = setPredictType(lrn2, "prob")
# train predict eval
mod = train(lrn2, multilabel.task)
pred = predict(mod, multilabel.task)
p = performance(pred)
expect_true(!is.na(p))
pmulti = getMultilabelBinaryPerformances(pred, list(mmce, auc))
expect_true(!any(is.na(pmulti)))
expect_true(is.matrix(pmulti))
# with newdata df
pred = predict(mod, newdata = multilabel.df)
p = performance(pred)
expect_true(!is.na(p))
pmulti = getMultilabelBinaryPerformances(pred, list(mmce, auc))
expect_true(is.matrix(pmulti))
expect_true(!any(is.na(pmulti)))
expect_equal(rownames(pmulti), getTaskTargetNames(multilabel.task))
expect_equal(colnames(pmulti), vcapply(list(mmce, auc), mlr:::measureAggrName))
# resample
r = holdout(lrn2, multilabel.task)
expect_true(!is.na(r$aggr))
pmulti = getMultilabelBinaryPerformances(pred, list(mmce, auc))
expect_true(is.matrix(pmulti))
expect_true(!any(is.na(pmulti)))
expect_equal(rownames(pmulti), getTaskTargetNames(multilabel.task))
expect_equal(colnames(pmulti), vcapply(list(mmce, auc), mlr:::measureAggrName))
lrn1 = makeLearner("classif.rpart", predict.type = "prob")
lrn2 = fun(lrn1, ...)
r = holdout(lrn2, multilabel.task)
expect_true(!is.na(r$aggr))
p = getPredictionProbabilities(r$pred)
expect_true(is.data.frame(p))
p = getPredictionProbabilities(r$pred, getTaskClassLevels(multilabel.task))
expect_true(is.data.frame(p))
lrn1 = makeLearner("classif.rpart")
lrn2 = fun(lrn1, ...)
lrn2 = setPredictType(lrn2, "prob")
r = holdout(lrn2, multilabel.task)
expect_true(!is.na(r$aggr))
# check some stuff for probs
cls = getTaskClassLevels(multilabel.task)
p = getPredictionProbabilities(r$pred)
expect_true(is.data.frame(p))
expect_equal(colnames(p), cls)
p = getPredictionProbabilities(r$pred, cls[1L])
expect_true(is.numeric(p))
# setThreshold
thresh = setThreshold(r$pred, threshold = c("y1" = 0.9, "y2" = 0.9))
expect_true(is.data.frame(thresh$data))
# now test that we can tune the thresholds
tr = tuneThreshold(r$pred, nsub = 2L, control = list(maxit = 2L))
expect_true(!is.na(tr$perf))
expect_equal(length(tr$th), length(getTaskClassLevels(multilabel.task)))
# Learner with Impute-Preprocessing
lrn1 = makeLearner("classif.rpart")
lrn2 = fun(lrn1, ...)
lrn2 = makeImputeWrapper(lrn2, classes = list(integer = imputeMedian(), numeric = imputeMedian(), factor = imputeConstant("Const")))
multilabel.df2 = multilabel.df
multilabel.df2[c(2, 10, 14), c(1, 5)] = NA
multilabel.task2 = makeMultilabelTask("multilabel", data = multilabel.df2, target = multilabel.target)
mod = train(lrn2, multilabel.task2)
pred = predict(mod, multilabel.task2)
p = performance(pred)
expect_true(!is.na(p))
# Learner with Hyperparameters
lrn1 = makeLearner("classif.rpart", par.vals = list(minsplit = 10, cp = 0.005))
lrn2 = fun(lrn1)
mod = train(lrn2, multilabel.task)
pred = predict(mod, multilabel.task)
p = performance(pred)
expect_true(!is.na(p))
# 3 targets
three.target.df = getTaskData(multilabel.task)
three.target.df$y3 = three.target.df$y2
multilabel3t.task = makeMultilabelTask(data = three.target.df, target = c("y1", "y2", "y3"))
mod = train(lrn2, multilabel3t.task)
pred = predict(mod, multilabel3t.task)
p = performance(pred)
expect_true(!is.na(p))
pmulti = getMultilabelBinaryPerformances(pred, list(mmce))
expect_true(!any(is.na(pmulti)))
# check order
args = list(...)
if (!is.null(args$order)) {
lrn2 = fun(lrn1, ...)
expect_error(train(lrn2, multilabel3t.task), "Must be equal to set|Must be a set equal")
}
})
}
testMultilabelWrapper(makeMultilabelBinaryRelevanceWrapper)
testMultilabelWrapper(makeMultilabelDBRWrapper)
testMultilabelWrapper(makeMultilabelClassifierChainsWrapper)
testMultilabelWrapper(makeMultilabelNestedStackingWrapper)
testMultilabelWrapper(makeMultilabelStackingWrapper)
# check order
testMultilabelWrapper(makeMultilabelClassifierChainsWrapper, order = c("y2", "y1"))
testMultilabelWrapper(makeMultilabelNestedStackingWrapper, order = c("y2", "y1"))
|
