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
|
test_that("tuneMBO", {
requirePackagesOrSkip("rgenoud", default.method = "load")
n.des = 8
n.iter = 2
res = makeResampleDesc("Holdout")
# tuning problem 1
ps1 = makeParamSet(
makeNumericParam("cp", lower = 0, upper = 1),
makeIntegerParam("minsplit", lower = 1, upper = 20)
)
lrn1 = makeLearner("classif.rpart")
# tuning problem 2
ps2 = makeParamSet(
makeNumericParam("sigma", lower = -10, upper = -1, trafo = function(x) 2^x)
)
lrn2 = makeLearner("classif.ksvm")
des2 = generateDesign(n.des, ps2)
# surrogate learner
sur.lrn = makeLearner("regr.lm", predict.type = "se")
# Problem 1 with manually defined mbo.ctrl and sur.lrn
mbo.ctrl = mlrMBO::makeMBOControl()
mbo.ctrl = mlrMBO::setMBOControlTermination(mbo.ctrl, iters = 2)
mbo.ctrl = mlrMBO::setMBOControlInfill(mbo.ctrl, crit = mlrMBO::crit.ei)
ctrl = makeTuneControlMBO(learner = sur.lrn, mbo.control = mbo.ctrl)
tr = tuneParams(lrn1, multiclass.task, res, par.set = ps1, control = ctrl,
measures = acc)
expect_equal(getOptPathLength(tr$opt.path), n.des + n.iter)
expect_equal(dim(as.data.frame(tr$opt.path)), c(n.des + n.iter, 2 + 1 + 4))
expect_number(tr$y, lower = 0.8, upper = 1)
# Problem 2 with manually given initial design
ctrl = makeTuneControlMBO(mbo.design = des2, budget = n.iter + n.des)
tr = tuneParams(lrn2, multiclass.task, res, par.set = ps2, control = ctrl)
expect_equal(getOptPathLength(tr$opt.path), n.des + n.iter)
expect_true(is.list(tr$x) && all(names(tr$x) == "sigma"))
expect_true(tr$x$sigma > 0)
df.op.trafo = as.data.frame(trafoOptPath(tr$opt.path))
expect_true(all(df.op.trafo$sigma > 0))
})
test_that("tuneMBO works with tune.threshold", {
# there was a bug here, see issue #2002
lrn = makeLearner("classif.rpart", predict.type = "prob")
ps = makeParamSet(
makeNumericParam("cp", lower = 0.1, upper = 1)
)
sur.lrn = makeLearner("regr.lm", predict.type = "se")
ctrl = makeTuneControlMBO(budget = 4L, tune.threshold = TRUE,
mbo.design = generateDesign(3, par.set = ps), learner = sur.lrn)
lrn = makeTuneWrapper(lrn, hout, par.set = ps, control = ctrl)
mod = train(lrn, sonar.task)
expect_number(mod$learner.model$opt.result$threshold, lower = 0, upper = 1)
})
|
