mlr3 是一个用于机器学习的R语言包。它是mlr包的下一代版本，旨在提供更强大、更灵活的机器学习功能。mlr3 提供了一种模块化的框架，使用户可以轻松地执行各种机器学习任务。
mlr3verse 是一组用于机器学习的R语言包的集合，旨在扩展mlr3框架的功能。这个集合包括了一系列用于数据预处理、特征选择、模型调优和评估的包，为用户提供了更多的工具和功能来进行机器学习任务。mlr3verse 的包括 mlr3 、mlr3learners 、mlr3pipelines、mlr3tuning、mlr3viz 等，每个包都有其特定的功能和用途，用户可以根据自己的需求选择适合的包来完成各种机器学习任务。
mlr3 中需要将数据封装为 task 类，然后再进行机器学习的相关操作
library(mlr3verse)
tsk_penguins <- as_task_classif(palmerpenguins::penguins,
target = "species",
id = "penguins"
)
tsk_penguins<TaskClassif:penguins> (344 x 8)
* Target: species
* Properties: multiclass
* Features (7):
- int (3): body_mass_g, flipper_length_mm, year
- dbl (2): bill_depth_mm, bill_length_mm
- fct (2): island, sex# 查看列结构
tsk_penguins$col_infoKey: <id>
id type levels label fix_factor_levels
<char> <char> <list> <char> <lgcl>
1: ..row_id integer [NULL] <NA> FALSE
2: bill_depth_mm numeric [NULL] <NA> FALSE
3: bill_length_mm numeric [NULL] <NA> FALSE
4: body_mass_g integer [NULL] <NA> FALSE
5: flipper_length_mm integer [NULL] <NA> FALSE
6: island factor Biscoe,Dream,Torgersen <NA> FALSE
7: sex factor female,male <NA> FALSE
8: species factor Adelie,Chinstrap,Gentoo <NA> FALSE
9: year integer [NULL] <NA> FALSE# 查看数据
tsk_penguins$data(1:5) species bill_depth_mm bill_length_mm body_mass_g flipper_length_mm island
<fctr> <num> <num> <int> <int> <fctr>
1: Adelie 18.7 39.1 3750 181 Torgersen
2: Adelie 17.4 39.5 3800 186 Torgersen
3: Adelie 18.0 40.3 3250 195 Torgersen
4: Adelie NA NA NA NA Torgersen
5: Adelie 19.3 36.7 3450 193 Torgersen
sex year
<fctr> <int>
1: male 2007
2: female 2007
3: female 2007
4: <NA> 2007
5: female 2007# 查看缺失值
tsk_penguins$missings() species bill_depth_mm bill_length_mm body_mass_g
0 2 2 2
flipper_length_mm island sex year
2 0 11 0 tsk_penguins$formula()species ~ .
NULLmlr3的数据预处理是通过mlr3pipelines实现的。mlr3pipelines中的数据预处理的所有步骤都采用了字符来表示，比较难记忆，可以在官网查看说明或通过mlr_pipeops来查看可用操作及说明
library(DT)
datatable(as.data.table(mlr_pipeops), options = list(
language = list(url = "https://cdn.datatables.net/plug-ins/1.10.11/i18n/Chinese.json"),
pageLength = 5
))从上表可以看到，mlr3verse 提供了丰富的数据处理方法。使用mlr3内置的iris等任务，我们抽取部分常用的数据预处理流程进行示例
task <- tsk("iris")
task<TaskClassif:iris> (150 x 5): Iris Flowers
* Target: Species
* Properties: multiclass
* Features (4):
- dbl (4): Petal.Length, Petal.Width, Sepal.Length, Sepal.Widthmlr3pipelines 中所有的数据预处理步骤都是通过po函数实现，每一个 po 除了自己的数据预处理参数之外，可以额外指定一个id作为该步骤的唯一名称，方便之后定位该步骤。
affect_columns参数可以控制该步骤影响的列，可以通过selector_*族函数来快捷的选择列
selector_*族函数
selector_all()：选择所有变量。
selector_none()：不选择任何变量。
selector_type(types)：按照变量类型选择，比如字符型、数值型等。
selector_grep(pattern, ignore.case = FALSE, perl = FALSE, fixed = FALSE)：使用正则表达式选择变量。
selector_name(feature_names, assert_present = FALSE)：按照变量名选择。
selector_invert(selector)：反选变量，即删除指定的变量。
selector_intersect(selector_x, selector_y)：选择两个选择器的交集。
selector_union(selector_x, selector_y)：选择两个选择器的并集。
selector_setdiff(selector_x, selector_y)：选择两个选择器的差集。
selector_missing()：选择存在缺失值的变量。
selector_cardinality_greater_than(min_cardinality)：选择分类特征基数（唯一值的数量）大于某个值的变量。
# 选择预处理步骤
pos <- po(
"scale",
center = T, # 中心化
scale = F, # 标准化
affect_columns = selector_name(c("Petal.Length", "Petal.Width", "Sepal.Length")), # 作用的变量
id = "scale_iris" # 为该步骤命名
)
posPipeOp: <scale_iris> (not trained)
values: <center=TRUE, scale=FALSE, robust=FALSE, affect_columns=<Selector>>
Input channels <name [train type, predict type]>:
input [Task,Task]
Output channels <name [train type, predict type]>:
output [Task,Task]# 执行并提取处理后数据
pos$train(list(task))[[1]]$data() Species Petal.Length Petal.Width Sepal.Length Sepal.Width
<fctr> <num> <num> <num> <num>
1: setosa -2.358 -0.9993333 -0.74333333 3.5
2: setosa -2.358 -0.9993333 -0.94333333 3.0
3: setosa -2.458 -0.9993333 -1.14333333 3.2
4: setosa -2.258 -0.9993333 -1.24333333 3.1
5: setosa -2.358 -0.9993333 -0.84333333 3.6
---
146: virginica 1.442 1.1006667 0.85666667 3.0
147: virginica 1.242 0.7006667 0.45666667 2.5
148: virginica 1.442 0.8006667 0.65666667 3.0
149: virginica 1.642 1.1006667 0.35666667 3.4
150: virginica 1.342 0.6006667 0.05666667 3.0imputelearner：算法插补
imputemean：均值
imputemedian：中位数
imputeconstant：常数
imoutehist：直方图
imputemode：众数插补
imputesample：随机
task <- tsk("pima")
task$missings()diabetes age glucose insulin mass pedigree pregnant pressure
0 0 5 374 11 0 0 35
triceps
227 # 决策树插补
po <- po("imputemean")
new_task <- po$train(list(task = task))[[1]]
new_task$missings()diabetes age pedigree pregnant glucose insulin mass pressure
0 0 0 0 0 0 0 0
triceps
0 task <- tsk("pima")
task$missings()diabetes age glucose insulin mass pedigree pregnant pressure
0 0 5 374 11 0 0 35
triceps
227 # 决策树插补
po <- po("imputelearner", lrn("regr.rpart"))
new_task <- po$train(list(task = task))[[1]]
new_task$missings()diabetes age pedigree pregnant glucose insulin mass pressure
0 0 0 0 0 0 0 0
triceps
0 # task = mlr3::tsk("mtcars")
# filter = flt("find_correlation")
# filter$calculate(task)
# as.data.table(filter)
library(dplyr)
task <- tsk("mtcars")
pos <-
# 去除高度相关的列
po("filter",
filter = mlr3filters::flt("find_correlation"),
filter.cutoff = 0.1
) %>>%
# 去掉常数
po("removeconstants") %>>%
# 去掉方差较小变量
po("filter",
filter = mlr3filters::flt("variance"),
filter.frac = 0.5
)
pos$train(task)[[1]]$data() mpg carb disp hp qsec
<num> <num> <num> <num> <num>
1: 21.0 4 160.0 110 16.46
2: 21.0 4 160.0 110 17.02
3: 22.8 1 108.0 93 18.61
4: 21.4 1 258.0 110 19.44
5: 18.7 2 360.0 175 17.02
6: 18.1 1 225.0 105 20.22
7: 14.3 4 360.0 245 15.84
8: 24.4 2 146.7 62 20.00
9: 22.8 2 140.8 95 22.90
10: 19.2 4 167.6 123 18.30
11: 17.8 4 167.6 123 18.90
12: 16.4 3 275.8 180 17.40
13: 17.3 3 275.8 180 17.60
14: 15.2 3 275.8 180 18.00
15: 10.4 4 472.0 205 17.98
16: 10.4 4 460.0 215 17.82
17: 14.7 4 440.0 230 17.42
18: 32.4 1 78.7 66 19.47
19: 30.4 2 75.7 52 18.52
20: 33.9 1 71.1 65 19.90
21: 21.5 1 120.1 97 20.01
22: 15.5 2 318.0 150 16.87
23: 15.2 2 304.0 150 17.30
24: 13.3 4 350.0 245 15.41
25: 19.2 2 400.0 175 17.05
26: 27.3 1 79.0 66 18.90
27: 26.0 2 120.3 91 16.70
28: 30.4 2 95.1 113 16.90
29: 15.8 4 351.0 264 14.50
30: 19.7 6 145.0 175 15.50
31: 15.0 8 301.0 335 14.60
32: 21.4 2 121.0 109 18.60
mpg carb disp hp qsecdata <- data.table::data.table(x = factor(letters[1:3]), y = factor(letters[1:3]))
task <- as_task_classif(data, target = "y")
poe <- po("encode", method = "one-hot")
# 默认 "one-hot"
poe$train(list(task))[[1]]$data() y x.a x.b x.c
<fctr> <num> <num> <num>
1: a 1 0 0
2: b 0 1 0
3: c 0 0 1poe$param_set$values$method <- "treatment"
poe$train(list(task))[[1]]$data() y x.b x.c
<fctr> <num> <num>
1: a 0 0
2: b 1 0
3: c 0 1ratio：和参考类别相比的倍数;
reference：设置参考类别;
adjust：选择过采样还是欠采样
shuffle：是否对结果打乱顺序，默认TRUE
data(hacide, package = "ROSE")
table(hacide.train$cls)
0 1
980 20 task <- as_task_classif(hacide.train, target = "cls")
pos <- po("classbalancing",
ratio = 1,
reference = "major",
adjust = "all",
shuffle = T
)
blanced <- pos$train(list(task))[[1]]$data()
table(blanced$cls)
0 1
980 980 task <- tsk("mtcars")
pos <- po("quantilebin", numsplits = 10, affect_columns = selector_name(c("disp", "hp")))
pos$train(list(task))[[1]]$data() %>% head() mpg disp hp am carb cyl drat gear qsec vs wt
<num> <ord> <ord> <num> <num> <num> <num> <num> <num> <num> <num>
1: 21.0 (142,160] (106,110] 1 4 6 3.90 4 16.46 0 2.620
2: 21.0 (142,160] (106,110] 1 4 6 3.90 4 17.02 0 2.875
3: 22.8 (80.6,120] (66,93.4] 1 1 4 3.85 4 18.61 1 2.320
4: 21.4 (196,276] (106,110] 0 1 6 3.08 3 19.44 1 3.215
5: 18.7 (351,396] (165,178] 0 2 8 3.15 3 17.02 0 3.440
6: 18.1 (196,276] (93.4,106] 0 1 6 2.76 3 20.22 1 3.460查看 mlr3 支持的算法
mlr_learners<DictionaryLearner> with 51 stored values
Keys: classif.cv_glmnet, classif.debug, classif.featureless,
classif.glmnet, classif.kknn, classif.lda, classif.log_reg,
classif.multinom, classif.naive_bayes, classif.nnet, classif.qda,
classif.ranger, classif.rpart, classif.svm, classif.xgboost,
clust.agnes, clust.ap, clust.bico, clust.birch, clust.cmeans,
clust.cobweb, clust.dbscan, clust.dbscan_fpc, clust.diana, clust.em,
clust.fanny, clust.featureless, clust.ff, clust.hclust,
clust.hdbscan, clust.kkmeans, clust.kmeans, clust.MBatchKMeans,
clust.mclust, clust.meanshift, clust.optics, clust.pam,
clust.SimpleKMeans, clust.xmeans, regr.cv_glmnet, regr.debug,
regr.featureless, regr.glmnet, regr.kknn, regr.km, regr.lm,
regr.nnet, regr.ranger, regr.rpart, regr.svm, regr.xgboost创建学习器
learner <- lrn("classif.rpart")
learner<LearnerClassifRpart:classif.rpart>: Classification Tree
* Model: -
* Parameters: xval=0
* Packages: mlr3, rpart
* Predict Types: [response], prob
* Feature Types: logical, integer, numeric, factor, ordered
* Properties: importance, missings, multiclass, selected_features,
twoclass, weights查看学习器支持的超参数
learner$param_set<ParamSet(10)>
id class lower upper nlevels default value
<char> <char> <num> <num> <num> <list> <list>
1: cp ParamDbl 0 1 Inf 0.01 [NULL]
2: keep_model ParamLgl NA NA 2 FALSE [NULL]
3: maxcompete ParamInt 0 Inf Inf 4 [NULL]
4: maxdepth ParamInt 1 30 30 30 [NULL]
5: maxsurrogate ParamInt 0 Inf Inf 5 [NULL]
6: minbucket ParamInt 1 Inf Inf <NoDefault[0]> [NULL]
7: minsplit ParamInt 1 Inf Inf 20 [NULL]
8: surrogatestyle ParamInt 0 1 2 0 [NULL]
9: usesurrogate ParamInt 0 2 3 2 [NULL]
10: xval ParamInt 0 Inf Inf 10 0设置学习器参数
learner <- lrn("classif.rpart", xval = 0, cp = 0.001)stratify设置分层抽样
task <- tsk("penguins") # 使用内置数据集
split <- partition(task, ratio = 0.6)learner$train(task, row_ids = split$train)prediction <- learner$predict(task, row_ids = split$test)
print(prediction)<PredictionClassif> for 138 observations:
row_ids truth response
3 Adelie Adelie
4 Adelie Adelie
6 Adelie Adelie
--- --- ---
331 Chinstrap Adelie
338 Chinstrap Chinstrap
343 Chinstrap Chinstrapprediction$confusion truth
response Adelie Chinstrap Gentoo
Adelie 64 3 0
Chinstrap 4 21 1
Gentoo 0 0 45autoplot(prediction)
# 查看支持的指标
# msrs()
measures <- msrs(c("classif.acc", "classif.ce"))
prediction$score(measures)classif.acc classif.ce
0.94202899 0.05797101 使用 benchmark 可以同时进行多个任务、多个模型、多重抽样方法的模型比较
design <- benchmark_grid(
tasks = tsks(c("spam", "german_credit", "sonar")),
learners = lrns(c("classif.ranger", "classif.rpart", "classif.featureless"), predict_type = "prob"),
resamplings = rsmps(c("holdout", "cv"))
)
print(design) task learner resampling
<char> <char> <char>
1: spam classif.ranger holdout
2: spam classif.ranger cv
3: spam classif.rpart holdout
4: spam classif.rpart cv
5: spam classif.featureless holdout
6: spam classif.featureless cv
7: german_credit classif.ranger holdout
8: german_credit classif.ranger cv
9: german_credit classif.rpart holdout
10: german_credit classif.rpart cv
11: german_credit classif.featureless holdout
12: german_credit classif.featureless cv
13: sonar classif.ranger holdout
14: sonar classif.ranger cv
15: sonar classif.rpart holdout
16: sonar classif.rpart cv
17: sonar classif.featureless holdout
18: sonar classif.featureless cvHoldout和CV都是用于评估机器学习模型性能的方法，但两者之间存在一些关键差异。
Holdout
Holdout方法是将数据集划分为训练集和测试集，其中训练集用于训练模型，测试集用于评估模型性能。训练集和测试集的大小通常是固定的，例如70%的训练集和30%的测试集。
Holdout方法简单易用，但存在以下缺点：
训练集和测试集的划分方式可能会影响模型性能评估结果。例如，如果训练集和测试集的分布不一致，则模型性能评估结果可能不准确。
Holdout方法只使用了一部分数据来训练模型，因此模型性能评估结果可能不够可靠。
CV
CV方法将数据集划分为多个子集，每个子集轮流作为训练集和测试集。这样可以使每个数据点都有机会被用作训练集和测试集，从而提高模型性能评估结果的可靠性。
常用的CV方法包括k折交叉验证和留一交叉验证。k折交叉验证将数据集划分为k个子集，每个子集轮流作为测试集，其余k-1个子集作为训练集。留一交叉验证将数据集划分为n个子集，其中每个子集包含一个数据点，每个数据点单独作为测试集，其余n-1个数据点作为训练集。
CV方法比Holdout方法更复杂，但具有以下优点：
CV方法可以充分利用数据，提高模型性能评估结果的可靠性。
CV方法可以用于选择最佳的超参数。
bmr <- benchmark(design, store_models = T)INFO [16:55:54.000] [mlr3] Running benchmark with 99 resampling iterations
INFO [16:55:54.071] [mlr3] Applying learner 'classif.ranger' on task 'spam' (iter 1/1)
INFO [16:55:55.746] [mlr3] Applying learner 'classif.ranger' on task 'spam' (iter 1/10)
INFO [16:55:57.779] [mlr3] Applying learner 'classif.ranger' on task 'spam' (iter 2/10)
INFO [16:55:59.510] [mlr3] Applying learner 'classif.ranger' on task 'spam' (iter 3/10)
INFO [16:56:01.265] [mlr3] Applying learner 'classif.ranger' on task 'spam' (iter 4/10)
INFO [16:56:03.834] [mlr3] Applying learner 'classif.ranger' on task 'spam' (iter 5/10)
INFO [16:56:06.207] [mlr3] Applying learner 'classif.ranger' on task 'spam' (iter 6/10)
INFO [16:56:09.214] [mlr3] Applying learner 'classif.ranger' on task 'spam' (iter 7/10)
INFO [16:56:12.729] [mlr3] Applying learner 'classif.ranger' on task 'spam' (iter 8/10)
INFO [16:56:15.533] [mlr3] Applying learner 'classif.ranger' on task 'spam' (iter 9/10)
INFO [16:56:18.217] [mlr3] Applying learner 'classif.ranger' on task 'spam' (iter 10/10)
INFO [16:56:21.763] [mlr3] Applying learner 'classif.rpart' on task 'spam' (iter 1/1)
INFO [16:56:21.850] [mlr3] Applying learner 'classif.rpart' on task 'spam' (iter 1/10)
INFO [16:56:21.958] [mlr3] Applying learner 'classif.rpart' on task 'spam' (iter 2/10)
INFO [16:56:22.066] [mlr3] Applying learner 'classif.rpart' on task 'spam' (iter 3/10)
INFO [16:56:22.176] [mlr3] Applying learner 'classif.rpart' on task 'spam' (iter 4/10)
INFO [16:56:22.282] [mlr3] Applying learner 'classif.rpart' on task 'spam' (iter 5/10)
INFO [16:56:22.389] [mlr3] Applying learner 'classif.rpart' on task 'spam' (iter 6/10)
INFO [16:56:22.492] [mlr3] Applying learner 'classif.rpart' on task 'spam' (iter 7/10)
INFO [16:56:22.583] [mlr3] Applying learner 'classif.rpart' on task 'spam' (iter 8/10)
INFO [16:56:22.674] [mlr3] Applying learner 'classif.rpart' on task 'spam' (iter 9/10)
INFO [16:56:22.771] [mlr3] Applying learner 'classif.rpart' on task 'spam' (iter 10/10)
INFO [16:56:22.865] [mlr3] Applying learner 'classif.featureless' on task 'spam' (iter 1/1)
INFO [16:56:22.889] [mlr3] Applying learner 'classif.featureless' on task 'spam' (iter 1/10)
INFO [16:56:22.911] [mlr3] Applying learner 'classif.featureless' on task 'spam' (iter 2/10)
INFO [16:56:22.932] [mlr3] Applying learner 'classif.featureless' on task 'spam' (iter 3/10)
INFO [16:56:22.951] [mlr3] Applying learner 'classif.featureless' on task 'spam' (iter 4/10)
INFO [16:56:22.973] [mlr3] Applying learner 'classif.featureless' on task 'spam' (iter 5/10)
INFO [16:56:22.995] [mlr3] Applying learner 'classif.featureless' on task 'spam' (iter 6/10)
INFO [16:56:23.017] [mlr3] Applying learner 'classif.featureless' on task 'spam' (iter 7/10)
INFO [16:56:23.040] [mlr3] Applying learner 'classif.featureless' on task 'spam' (iter 8/10)
INFO [16:56:23.066] [mlr3] Applying learner 'classif.featureless' on task 'spam' (iter 9/10)
INFO [16:56:23.090] [mlr3] Applying learner 'classif.featureless' on task 'spam' (iter 10/10)
INFO [16:56:23.114] [mlr3] Applying learner 'classif.ranger' on task 'german_credit' (iter 1/1)
INFO [16:56:23.404] [mlr3] Applying learner 'classif.ranger' on task 'german_credit' (iter 1/10)
INFO [16:56:23.738] [mlr3] Applying learner 'classif.ranger' on task 'german_credit' (iter 2/10)
INFO [16:56:24.050] [mlr3] Applying learner 'classif.ranger' on task 'german_credit' (iter 3/10)
INFO [16:56:24.244] [mlr3] Applying learner 'classif.ranger' on task 'german_credit' (iter 4/10)
INFO [16:56:24.447] [mlr3] Applying learner 'classif.ranger' on task 'german_credit' (iter 5/10)
INFO [16:56:24.651] [mlr3] Applying learner 'classif.ranger' on task 'german_credit' (iter 6/10)
INFO [16:56:24.857] [mlr3] Applying learner 'classif.ranger' on task 'german_credit' (iter 7/10)
INFO [16:56:25.078] [mlr3] Applying learner 'classif.ranger' on task 'german_credit' (iter 8/10)
INFO [16:56:25.279] [mlr3] Applying learner 'classif.ranger' on task 'german_credit' (iter 9/10)
INFO [16:56:25.953] [mlr3] Applying learner 'classif.ranger' on task 'german_credit' (iter 10/10)
INFO [16:56:26.183] [mlr3] Applying learner 'classif.rpart' on task 'german_credit' (iter 1/1)
INFO [16:56:26.208] [mlr3] Applying learner 'classif.rpart' on task 'german_credit' (iter 1/10)
INFO [16:56:26.237] [mlr3] Applying learner 'classif.rpart' on task 'german_credit' (iter 2/10)
INFO [16:56:26.264] [mlr3] Applying learner 'classif.rpart' on task 'german_credit' (iter 3/10)
INFO [16:56:26.288] [mlr3] Applying learner 'classif.rpart' on task 'german_credit' (iter 4/10)
INFO [16:56:26.311] [mlr3] Applying learner 'classif.rpart' on task 'german_credit' (iter 5/10)
INFO [16:56:26.338] [mlr3] Applying learner 'classif.rpart' on task 'german_credit' (iter 6/10)
INFO [16:56:26.364] [mlr3] Applying learner 'classif.rpart' on task 'german_credit' (iter 7/10)
INFO [16:56:26.393] [mlr3] Applying learner 'classif.rpart' on task 'german_credit' (iter 8/10)
INFO [16:56:26.419] [mlr3] Applying learner 'classif.rpart' on task 'german_credit' (iter 9/10)
INFO [16:56:26.445] [mlr3] Applying learner 'classif.rpart' on task 'german_credit' (iter 10/10)
INFO [16:56:26.476] [mlr3] Applying learner 'classif.featureless' on task 'german_credit' (iter 1/1)
INFO [16:56:26.490] [mlr3] Applying learner 'classif.featureless' on task 'german_credit' (iter 1/10)
INFO [16:56:26.507] [mlr3] Applying learner 'classif.featureless' on task 'german_credit' (iter 2/10)
INFO [16:56:26.522] [mlr3] Applying learner 'classif.featureless' on task 'german_credit' (iter 3/10)
INFO [16:56:26.538] [mlr3] Applying learner 'classif.featureless' on task 'german_credit' (iter 4/10)
INFO [16:56:26.554] [mlr3] Applying learner 'classif.featureless' on task 'german_credit' (iter 5/10)
INFO [16:56:26.576] [mlr3] Applying learner 'classif.featureless' on task 'german_credit' (iter 6/10)
INFO [16:56:26.594] [mlr3] Applying learner 'classif.featureless' on task 'german_credit' (iter 7/10)
INFO [16:56:26.613] [mlr3] Applying learner 'classif.featureless' on task 'german_credit' (iter 8/10)
INFO [16:56:26.632] [mlr3] Applying learner 'classif.featureless' on task 'german_credit' (iter 9/10)
INFO [16:56:26.646] [mlr3] Applying learner 'classif.featureless' on task 'german_credit' (iter 10/10)
INFO [16:56:26.661] [mlr3] Applying learner 'classif.ranger' on task 'sonar' (iter 1/1)
INFO [16:56:26.739] [mlr3] Applying learner 'classif.ranger' on task 'sonar' (iter 1/10)
INFO [16:56:26.825] [mlr3] Applying learner 'classif.ranger' on task 'sonar' (iter 2/10)
INFO [16:56:26.915] [mlr3] Applying learner 'classif.ranger' on task 'sonar' (iter 3/10)
INFO [16:56:27.006] [mlr3] Applying learner 'classif.ranger' on task 'sonar' (iter 4/10)
INFO [16:56:27.105] [mlr3] Applying learner 'classif.ranger' on task 'sonar' (iter 5/10)
INFO [16:56:27.204] [mlr3] Applying learner 'classif.ranger' on task 'sonar' (iter 6/10)
INFO [16:56:27.304] [mlr3] Applying learner 'classif.ranger' on task 'sonar' (iter 7/10)
INFO [16:56:27.397] [mlr3] Applying learner 'classif.ranger' on task 'sonar' (iter 8/10)
INFO [16:56:27.503] [mlr3] Applying learner 'classif.ranger' on task 'sonar' (iter 9/10)
INFO [16:56:27.592] [mlr3] Applying learner 'classif.ranger' on task 'sonar' (iter 10/10)
INFO [16:56:27.681] [mlr3] Applying learner 'classif.rpart' on task 'sonar' (iter 1/1)
INFO [16:56:27.703] [mlr3] Applying learner 'classif.rpart' on task 'sonar' (iter 1/10)
INFO [16:56:27.727] [mlr3] Applying learner 'classif.rpart' on task 'sonar' (iter 2/10)
INFO [16:56:27.750] [mlr3] Applying learner 'classif.rpart' on task 'sonar' (iter 3/10)
INFO [16:56:27.775] [mlr3] Applying learner 'classif.rpart' on task 'sonar' (iter 4/10)
INFO [16:56:27.802] [mlr3] Applying learner 'classif.rpart' on task 'sonar' (iter 5/10)
INFO [16:56:27.832] [mlr3] Applying learner 'classif.rpart' on task 'sonar' (iter 6/10)
INFO [16:56:27.864] [mlr3] Applying learner 'classif.rpart' on task 'sonar' (iter 7/10)
INFO [16:56:27.894] [mlr3] Applying learner 'classif.rpart' on task 'sonar' (iter 8/10)
INFO [16:56:27.922] [mlr3] Applying learner 'classif.rpart' on task 'sonar' (iter 9/10)
INFO [16:56:27.951] [mlr3] Applying learner 'classif.rpart' on task 'sonar' (iter 10/10)
INFO [16:56:27.981] [mlr3] Applying learner 'classif.featureless' on task 'sonar' (iter 1/1)
INFO [16:56:27.995] [mlr3] Applying learner 'classif.featureless' on task 'sonar' (iter 1/10)
INFO [16:56:28.013] [mlr3] Applying learner 'classif.featureless' on task 'sonar' (iter 2/10)
INFO [16:56:28.031] [mlr3] Applying learner 'classif.featureless' on task 'sonar' (iter 3/10)
INFO [16:56:28.047] [mlr3] Applying learner 'classif.featureless' on task 'sonar' (iter 4/10)
INFO [16:56:28.066] [mlr3] Applying learner 'classif.featureless' on task 'sonar' (iter 5/10)
INFO [16:56:28.083] [mlr3] Applying learner 'classif.featureless' on task 'sonar' (iter 6/10)
INFO [16:56:28.098] [mlr3] Applying learner 'classif.featureless' on task 'sonar' (iter 7/10)
INFO [16:56:28.112] [mlr3] Applying learner 'classif.featureless' on task 'sonar' (iter 8/10)
INFO [16:56:28.128] [mlr3] Applying learner 'classif.featureless' on task 'sonar' (iter 9/10)
INFO [16:56:28.146] [mlr3] Applying learner 'classif.featureless' on task 'sonar' (iter 10/10)
INFO [16:56:28.188] [mlr3] Finished benchmarkmeasures <- msrs(c("classif.acc", "classif.mcc"))
bmr$aggregate(measures) nr task_id learner_id resampling_id iters classif.acc
<int> <char> <char> <char> <int> <num>
1: 1 spam classif.ranger holdout 1 0.9445893
2: 2 spam classif.ranger cv 10 0.9511002
3: 3 spam classif.rpart holdout 1 0.8943937
4: 4 spam classif.rpart cv 10 0.8950240
5: 5 spam classif.featureless holdout 1 0.6043025
6: 6 spam classif.featureless cv 10 0.6059559
7: 7 german_credit classif.ranger holdout 1 0.7777778
8: 8 german_credit classif.ranger cv 10 0.7630000
9: 9 german_credit classif.rpart holdout 1 0.7417417
10: 10 german_credit classif.rpart cv 10 0.7430000
11: 11 german_credit classif.featureless holdout 1 0.7267267
12: 12 german_credit classif.featureless cv 10 0.7000000
13: 13 sonar classif.ranger holdout 1 0.8550725
14: 14 sonar classif.ranger cv 10 0.8350000
15: 15 sonar classif.rpart holdout 1 0.7391304
16: 16 sonar classif.rpart cv 10 0.7061905
17: 17 sonar classif.featureless holdout 1 0.3913043
18: 18 sonar classif.featureless cv 10 0.5330952
classif.mcc
<num>
1: 0.8839099
2: 0.8974370
3: 0.7782326
4: 0.7788838
5: 0.0000000
6: 0.0000000
7: 0.3920031
8: 0.3817242
9: 0.2930620
10: 0.3407961
11: 0.0000000
12: 0.0000000
13: 0.6930077
14: 0.6792771
15: 0.4603609
16: 0.4064497
17: 0.0000000
18: 0.0000000
Hidden columns: resample_resultlibrary(ggplot2)
autoplot(bmr)
机器学习模型在实际应用中，往往会遇到性能不佳的问题。机器学习的模型都有默认的超参数，但默认的超参数并不一定最适合你的模型，在这种情况下，就需要进行超参数调优。
mlr3包含自动调参的策略，自动调参需要指定以下信息：
搜索空间：指模型超参数取值的范围。
优化算法：指用于搜索最优解的算法。
评估方法：指用于评估模型性能的方法。
评价指标：指用于衡量模型性能的指标
调优示例
library(mlr3verse)
task <- tsk("pima")
learner <- lrn("classif.rpart")
# 查看算法支持的超参数
learner$param_set<ParamSet(10)>
id class lower upper nlevels default value
<char> <char> <num> <num> <num> <list> <list>
1: cp ParamDbl 0 1 Inf 0.01 [NULL]
2: keep_model ParamLgl NA NA 2 FALSE [NULL]
3: maxcompete ParamInt 0 Inf Inf 4 [NULL]
4: maxdepth ParamInt 1 30 30 30 [NULL]
5: maxsurrogate ParamInt 0 Inf Inf 5 [NULL]
6: minbucket ParamInt 1 Inf Inf <NoDefault[0]> [NULL]
7: minsplit ParamInt 1 Inf Inf 20 [NULL]
8: surrogatestyle ParamInt 0 1 2 0 [NULL]
9: usesurrogate ParamInt 0 2 3 2 [NULL]
10: xval ParamInt 0 Inf Inf 10 0设置调参空间
search_space <- ps(
cp = p_dbl(lower = 0.001, upper = 0.1), # 复杂度参数
minsplit = p_int(lower = 1, upper = 10)
)
search_space<ParamSet(2)>
id class lower upper nlevels default value
<char> <char> <num> <num> <num> <list> <list>
1: cp ParamDbl 0.001 0.1 Inf <NoDefault[0]> [NULL]
2: minsplit ParamInt 1.000 10.0 10 <NoDefault[0]> [NULL]设置重抽样方法和性能指标
cv <- rsmp("cv")
measures <- msrs(c("classif.ce", "time_train", "classif.acc"))有
library(mlr3tuning)
# 设置终止条件为10轮后停止，通过mlr_terminators可以查看支持的其他终止条件，比如run_time可以时长
instance <- tune(
tuner = tnr("grid_search", resolution = 5, batch_size = 2),
task = task,
learner = learner,
resampling = cv,
measure = measures,
search_space = search_space,
term_evals = 10
)INFO [16:56:30.873] [bbotk] Starting to optimize 2 parameter(s) with '<OptimizerBatchGridSearch>' and '<TerminatorEvals> [n_evals=10, k=0]'
INFO [16:56:30.891] [bbotk] Evaluating 2 configuration(s)
INFO [16:56:30.898] [mlr3] Running benchmark with 20 resampling iterations
INFO [16:56:30.905] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/10)
INFO [16:56:30.929] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 2/10)
INFO [16:56:30.958] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 3/10)
INFO [16:56:30.980] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 4/10)
INFO [16:56:31.006] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 5/10)
INFO [16:56:31.033] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 6/10)
INFO [16:56:31.061] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 7/10)
INFO [16:56:31.090] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 8/10)
INFO [16:56:31.117] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 9/10)
INFO [16:56:31.147] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 10/10)
INFO [16:56:31.173] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/10)
INFO [16:56:31.196] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 2/10)
INFO [16:56:31.220] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 3/10)
INFO [16:56:31.242] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 4/10)
INFO [16:56:31.264] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 5/10)
INFO [16:56:31.286] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 6/10)
INFO [16:56:31.306] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 7/10)
INFO [16:56:31.325] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 8/10)
INFO [16:56:31.344] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 9/10)
INFO [16:56:31.364] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 10/10)
INFO [16:56:31.385] [mlr3] Finished benchmark
INFO [16:56:31.549] [bbotk] Result of batch 1:
INFO [16:56:31.552] [bbotk] cp minsplit classif.ce time_train classif.acc warnings errors
INFO [16:56:31.552] [bbotk] 0.0010 3 0.3006835 0.006 0.6993165 0 0
INFO [16:56:31.552] [bbotk] 0.0505 8 0.2408578 0.005 0.7591422 0 0
INFO [16:56:31.552] [bbotk] runtime_learners uhash
INFO [16:56:31.552] [bbotk] 0.08 021bc8da-3580-4b5c-bfb1-3488b43f7b05
INFO [16:56:31.552] [bbotk] 0.08 d3483c94-59e0-497c-9608-d82e2d1d3e6f
INFO [16:56:31.556] [bbotk] Evaluating 2 configuration(s)
INFO [16:56:31.564] [mlr3] Running benchmark with 20 resampling iterations
INFO [16:56:31.586] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/10)
INFO [16:56:31.625] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 2/10)
INFO [16:56:31.650] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 3/10)
INFO [16:56:31.679] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 4/10)
INFO [16:56:31.709] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 5/10)
INFO [16:56:31.734] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 6/10)
INFO [16:56:31.752] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 7/10)
INFO [16:56:31.771] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 8/10)
INFO [16:56:31.790] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 9/10)
INFO [16:56:31.812] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 10/10)
INFO [16:56:31.839] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/10)
INFO [16:56:31.864] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 2/10)
INFO [16:56:31.890] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 3/10)
INFO [16:56:31.919] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 4/10)
INFO [16:56:31.949] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 5/10)
INFO [16:56:31.972] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 6/10)
INFO [16:56:31.995] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 7/10)
INFO [16:56:32.016] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 8/10)
INFO [16:56:32.039] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 9/10)
INFO [16:56:32.062] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 10/10)
INFO [16:56:32.086] [mlr3] Finished benchmark
INFO [16:56:32.226] [bbotk] Result of batch 2:
INFO [16:56:32.229] [bbotk] cp minsplit classif.ce time_train classif.acc warnings errors
INFO [16:56:32.229] [bbotk] 0.02575 1 0.2460526 0.006 0.7539474 0 0
INFO [16:56:32.229] [bbotk] 0.10000 3 0.2616883 0.005 0.7383117 0 0
INFO [16:56:32.229] [bbotk] runtime_learners uhash
INFO [16:56:32.229] [bbotk] 0.08 a2355435-5006-4b41-8120-635d920dd14a
INFO [16:56:32.229] [bbotk] 0.08 0bf4db31-4f73-4f97-a382-98bd2741895f
INFO [16:56:32.234] [bbotk] Evaluating 2 configuration(s)
INFO [16:56:32.243] [mlr3] Running benchmark with 20 resampling iterations
INFO [16:56:32.248] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/10)
INFO [16:56:32.280] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 2/10)
INFO [16:56:32.300] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 3/10)
INFO [16:56:32.320] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 4/10)
INFO [16:56:32.341] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 5/10)
INFO [16:56:32.364] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 6/10)
INFO [16:56:32.388] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 7/10)
INFO [16:56:32.410] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 8/10)
INFO [16:56:32.432] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 9/10)
INFO [16:56:32.457] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 10/10)
INFO [16:56:32.478] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/10)
INFO [16:56:32.498] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 2/10)
INFO [16:56:32.519] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 3/10)
INFO [16:56:32.539] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 4/10)
INFO [16:56:32.560] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 5/10)
INFO [16:56:32.583] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 6/10)
INFO [16:56:32.604] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 7/10)
INFO [16:56:32.630] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 8/10)
INFO [16:56:32.656] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 9/10)
INFO [16:56:32.678] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 10/10)
INFO [16:56:32.705] [mlr3] Finished benchmark
INFO [16:56:32.845] [bbotk] Result of batch 3:
INFO [16:56:32.847] [bbotk] cp minsplit classif.ce time_train classif.acc warnings errors
INFO [16:56:32.847] [bbotk] 0.07525 1 0.2460526 0.006 0.7539474 0 0
INFO [16:56:32.847] [bbotk] 0.10000 1 0.2616883 0.004 0.7383117 0 0
INFO [16:56:32.847] [bbotk] runtime_learners uhash
INFO [16:56:32.847] [bbotk] 0.10 f5e98795-bd80-4a58-9f5d-6ee2b461769c
INFO [16:56:32.847] [bbotk] 0.05 5a52f66b-cacb-49ba-bf53-d2266982baed
INFO [16:56:32.850] [bbotk] Evaluating 2 configuration(s)
INFO [16:56:32.856] [mlr3] Running benchmark with 20 resampling iterations
INFO [16:56:32.861] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/10)
INFO [16:56:32.882] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 2/10)
INFO [16:56:32.903] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 3/10)
INFO [16:56:32.926] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 4/10)
INFO [16:56:32.949] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 5/10)
INFO [16:56:32.971] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 6/10)
INFO [16:56:32.995] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 7/10)
INFO [16:56:33.019] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 8/10)
INFO [16:56:33.047] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 9/10)
INFO [16:56:33.071] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 10/10)
INFO [16:56:33.094] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/10)
INFO [16:56:33.131] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 2/10)
INFO [16:56:33.155] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 3/10)
INFO [16:56:33.179] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 4/10)
INFO [16:56:33.208] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 5/10)
INFO [16:56:33.235] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 6/10)
INFO [16:56:33.261] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 7/10)
INFO [16:56:33.285] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 8/10)
INFO [16:56:33.309] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 9/10)
INFO [16:56:33.333] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 10/10)
INFO [16:56:33.356] [mlr3] Finished benchmark
INFO [16:56:33.537] [bbotk] Result of batch 4:
INFO [16:56:33.540] [bbotk] cp minsplit classif.ce time_train classif.acc warnings errors
INFO [16:56:33.540] [bbotk] 0.00100 10 0.2654990 0.007 0.7345010 0 0
INFO [16:56:33.540] [bbotk] 0.07525 6 0.2460526 0.007 0.7539474 0 0
INFO [16:56:33.540] [bbotk] runtime_learners uhash
INFO [16:56:33.540] [bbotk] 0.11 e5b7ba78-65bb-47aa-af69-0d1f967ab722
INFO [16:56:33.540] [bbotk] 0.12 9b6cd101-b678-4913-a8fa-7942bfb99820
INFO [16:56:33.544] [bbotk] Evaluating 2 configuration(s)
INFO [16:56:33.553] [mlr3] Running benchmark with 20 resampling iterations
INFO [16:56:33.560] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/10)
INFO [16:56:33.588] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 2/10)
INFO [16:56:33.615] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 3/10)
INFO [16:56:33.639] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 4/10)
INFO [16:56:33.665] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 5/10)
INFO [16:56:33.690] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 6/10)
INFO [16:56:33.710] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 7/10)
INFO [16:56:33.732] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 8/10)
INFO [16:56:33.754] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 9/10)
INFO [16:56:33.779] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 10/10)
INFO [16:56:33.798] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/10)
INFO [16:56:33.818] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 2/10)
INFO [16:56:33.836] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 3/10)
INFO [16:56:33.853] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 4/10)
INFO [16:56:33.874] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 5/10)
INFO [16:56:33.895] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 6/10)
INFO [16:56:33.915] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 7/10)
INFO [16:56:33.936] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 8/10)
INFO [16:56:33.956] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 9/10)
INFO [16:56:33.975] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 10/10)
INFO [16:56:34.011] [mlr3] Finished benchmark
INFO [16:56:34.149] [bbotk] Result of batch 5:
INFO [16:56:34.151] [bbotk] cp minsplit classif.ce time_train classif.acc warnings errors
INFO [16:56:34.151] [bbotk] 0.0010 1 0.2994019 0.004 0.7005981 0 0
INFO [16:56:34.151] [bbotk] 0.0505 3 0.2408578 0.006 0.7591422 0 0
INFO [16:56:34.151] [bbotk] runtime_learners uhash
INFO [16:56:34.151] [bbotk] 0.06 8d4963c6-ee8b-4c70-9694-9e3e989bb0bf
INFO [16:56:34.151] [bbotk] 0.07 4f947ca3-81a7-46bc-b0f9-97b6e943d4c7
INFO [16:56:34.171] [bbotk] Finished optimizing after 10 evaluation(s)
INFO [16:56:34.172] [bbotk] Result:
INFO [16:56:34.175] [bbotk] cp minsplit learner_param_vals x_domain classif.ce time_train classif.acc
INFO [16:56:34.175] [bbotk] <num> <int> <list> <list> <num> <num> <num>
INFO [16:56:34.175] [bbotk] 0.0505 8 <list[3]> <list[2]> 0.2408578 0.005 0.7591422
INFO [16:56:34.175] [bbotk] 0.1000 1 <list[3]> <list[2]> 0.2616883 0.004 0.7383117
INFO [16:56:34.175] [bbotk] 0.0010 1 <list[3]> <list[2]> 0.2994019 0.004 0.7005981# 设置搜索方法：grid_search为网格搜索，random_search 为随机搜索
# 注意这里设置的resolution = 5，表示会基于cp和minsplit设置5*5的均匀网格搜索，正常会搜索25个组合的参数，但是由于前面我们设置了最大轮次为10轮，所以10次就结束了
instance<TuningInstanceBatchMultiCrit>
* State: Optimized
* Objective: <ObjectiveTuningBatch:classif.rpart_on_pima>
* Search Space:
id class lower upper nlevels
<char> <char> <num> <num> <num>
1: cp ParamDbl 0.001 0.1 Inf
2: minsplit ParamInt 1.000 10.0 10
* Terminator: <TerminatorEvals>
* Result:
cp minsplit classif.ce time_train classif.acc
<num> <int> <num> <num> <num>
1: 0.0505 8 0.2408578 0.005 0.7591422
2: 0.1000 1 0.2616883 0.004 0.7383117
3: 0.0010 1 0.2994019 0.004 0.7005981
* Archive:
cp minsplit classif.ce time_train classif.acc
<num> <int> <num> <num> <num>
1: 0.00100 3 0.3006835 0.006 0.6993165
2: 0.05050 8 0.2408578 0.005 0.7591422
3: 0.02575 1 0.2460526 0.006 0.7539474
4: 0.10000 3 0.2616883 0.005 0.7383117
5: 0.07525 1 0.2460526 0.006 0.7539474
6: 0.10000 1 0.2616883 0.004 0.7383117
7: 0.00100 10 0.2654990 0.007 0.7345010
8: 0.07525 6 0.2460526 0.007 0.7539474
9: 0.00100 1 0.2994019 0.004 0.7005981
10: 0.05050 3 0.2408578 0.006 0.7591422查看调参结果
instance<TuningInstanceBatchMultiCrit>
* State: Optimized
* Objective: <ObjectiveTuningBatch:classif.rpart_on_pima>
* Search Space:
id class lower upper nlevels
<char> <char> <num> <num> <num>
1: cp ParamDbl 0.001 0.1 Inf
2: minsplit ParamInt 1.000 10.0 10
* Terminator: <TerminatorEvals>
* Result:
cp minsplit classif.ce time_train classif.acc
<num> <int> <num> <num> <num>
1: 0.0505 8 0.2408578 0.005 0.7591422
2: 0.1000 1 0.2616883 0.004 0.7383117
3: 0.0010 1 0.2994019 0.004 0.7005981
* Archive:
cp minsplit classif.ce time_train classif.acc
<num> <int> <num> <num> <num>
1: 0.00100 3 0.3006835 0.006 0.6993165
2: 0.05050 8 0.2408578 0.005 0.7591422
3: 0.02575 1 0.2460526 0.006 0.7539474
4: 0.10000 3 0.2616883 0.005 0.7383117
5: 0.07525 1 0.2460526 0.006 0.7539474
6: 0.10000 1 0.2616883 0.004 0.7383117
7: 0.00100 10 0.2654990 0.007 0.7345010
8: 0.07525 6 0.2460526 0.007 0.7539474
9: 0.00100 1 0.2994019 0.004 0.7005981
10: 0.05050 3 0.2408578 0.006 0.7591422将训练好的参数应用于模型，重新训练数据
查看调整好的参数
instance$result_learner_param_vals[[1]]
[[1]]$xval
[1] 0
[[1]]$cp
[1] 0.0505
[[1]]$minsplit
[1] 8
[[2]]
[[2]]$xval
[1] 0
[[2]]$cp
[1] 0.1
[[2]]$minsplit
[1] 1
[[3]]
[[3]]$xval
[1] 0
[[3]]$cp
[1] 0.001
[[3]]$minsplit
[1] 1模型性能
instance$result_y classif.ce time_train classif.acc
<num> <num> <num>
1: 0.2408578 0.005 0.7591422
2: 0.2616883 0.004 0.7383117
3: 0.2994019 0.004 0.7005981将调优选择参数应用回模型
learner$param_set$values <- instance$result_learner_param_vals[[1]]
learner$train(task)
pred <- learner$predict(task)
pred$confusion truth
response pos neg
pos 150 58
neg 118 442pred$score(msr("classif.acc"))classif.acc
0.7708333 auto_learner可以直接返回最优的那个模型
task <- tsk("pima")
leanrer <- lrn("classif.rpart")
search_space <- ps(
cp = p_dbl(0.001, 0.1),
minsplit = p_int(1, 10)
)
cv <- rsmp("cv")
measures <- msr("classif.acc")
auto_learner <- auto_tuner(
tuner = tnr("random_search", batch_size = 2),
learner = learner,
resampling = cv,
measure = measures,
search_space = search_space,
term_evals = 10
)
auto_learner$train(task)INFO [16:56:34.529] [bbotk] Starting to optimize 2 parameter(s) with '<OptimizerBatchRandomSearch>' and '<TerminatorEvals> [n_evals=10, k=0]'
INFO [16:56:34.558] [bbotk] Evaluating 2 configuration(s)
INFO [16:56:34.566] [mlr3] Running benchmark with 20 resampling iterations
INFO [16:56:34.572] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/10)
INFO [16:56:34.594] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 2/10)
INFO [16:56:34.618] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 3/10)
INFO [16:56:34.637] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 4/10)
INFO [16:56:34.659] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 5/10)
INFO [16:56:34.682] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 6/10)
INFO [16:56:34.701] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 7/10)
INFO [16:56:34.722] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 8/10)
INFO [16:56:34.746] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 9/10)
INFO [16:56:34.766] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 10/10)
INFO [16:56:34.790] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/10)
INFO [16:56:34.812] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 2/10)
INFO [16:56:34.832] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 3/10)
INFO [16:56:34.853] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 4/10)
INFO [16:56:34.880] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 5/10)
INFO [16:56:34.914] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 6/10)
INFO [16:56:34.937] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 7/10)
INFO [16:56:34.959] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 8/10)
INFO [16:56:34.980] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 9/10)
INFO [16:56:35.003] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 10/10)
INFO [16:56:35.025] [mlr3] Finished benchmark
INFO [16:56:35.080] [bbotk] Result of batch 1:
INFO [16:56:35.082] [bbotk] cp minsplit classif.acc warnings errors runtime_learners
INFO [16:56:35.082] [bbotk] 0.09917854 7 0.7357826 0 0 0.12
INFO [16:56:35.082] [bbotk] 0.07838151 6 0.7448735 0 0 0.05
INFO [16:56:35.082] [bbotk] uhash
INFO [16:56:35.082] [bbotk] 0e27424c-e253-426a-979c-c5060ca88b8e
INFO [16:56:35.082] [bbotk] 005c77b1-69b0-40bd-9637-837ea8b6659c
INFO [16:56:35.092] [bbotk] Evaluating 2 configuration(s)
INFO [16:56:35.098] [mlr3] Running benchmark with 20 resampling iterations
INFO [16:56:35.105] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/10)
INFO [16:56:35.125] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 2/10)
INFO [16:56:35.149] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 3/10)
INFO [16:56:35.171] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 4/10)
INFO [16:56:35.189] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 5/10)
INFO [16:56:35.211] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 6/10)
INFO [16:56:35.232] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 7/10)
INFO [16:56:35.253] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 8/10)
INFO [16:56:35.272] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 9/10)
INFO [16:56:35.291] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 10/10)
INFO [16:56:35.311] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/10)
INFO [16:56:35.328] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 2/10)
INFO [16:56:35.347] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 3/10)
INFO [16:56:35.365] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 4/10)
INFO [16:56:35.382] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 5/10)
INFO [16:56:35.399] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 6/10)
INFO [16:56:35.416] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 7/10)
INFO [16:56:35.434] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 8/10)
INFO [16:56:35.451] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 9/10)
INFO [16:56:35.469] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 10/10)
INFO [16:56:35.488] [mlr3] Finished benchmark
INFO [16:56:35.550] [bbotk] Result of batch 2:
INFO [16:56:35.553] [bbotk] cp minsplit classif.acc warnings errors runtime_learners
INFO [16:56:35.553] [bbotk] 0.07918972 1 0.7448735 0 0 0.05
INFO [16:56:35.553] [bbotk] 0.08291213 1 0.7448735 0 0 0.06
INFO [16:56:35.553] [bbotk] uhash
INFO [16:56:35.553] [bbotk] aaaaf98d-c721-435d-a403-8b1468152cf5
INFO [16:56:35.553] [bbotk] 55c17ed0-8743-4685-973e-3819f2fab616
INFO [16:56:35.562] [bbotk] Evaluating 2 configuration(s)
INFO [16:56:35.570] [mlr3] Running benchmark with 20 resampling iterations
INFO [16:56:35.575] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/10)
INFO [16:56:35.599] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 2/10)
INFO [16:56:35.620] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 3/10)
INFO [16:56:35.642] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 4/10)
INFO [16:56:35.662] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 5/10)
INFO [16:56:35.728] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 6/10)
INFO [16:56:35.748] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 7/10)
INFO [16:56:35.768] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 8/10)
INFO [16:56:35.786] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 9/10)
INFO [16:56:35.806] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 10/10)
INFO [16:56:35.828] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/10)
INFO [16:56:35.849] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 2/10)
INFO [16:56:35.869] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 3/10)
INFO [16:56:35.895] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 4/10)
INFO [16:56:35.920] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 5/10)
INFO [16:56:35.945] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 6/10)
INFO [16:56:35.971] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 7/10)
INFO [16:56:35.999] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 8/10)
INFO [16:56:36.026] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 9/10)
INFO [16:56:36.051] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 10/10)
INFO [16:56:36.081] [mlr3] Finished benchmark
INFO [16:56:36.159] [bbotk] Result of batch 3:
INFO [16:56:36.162] [bbotk] cp minsplit classif.acc warnings errors runtime_learners
INFO [16:56:36.162] [bbotk] 0.08689330 6 0.7357826 0 0 0.13
INFO [16:56:36.162] [bbotk] 0.05292542 2 0.7487697 0 0 0.06
INFO [16:56:36.162] [bbotk] uhash
INFO [16:56:36.162] [bbotk] e7fa3cc7-71cd-4acd-84e2-f6b58e2edc7b
INFO [16:56:36.162] [bbotk] e5f2a260-03b7-40e2-a4ca-7374ff1c8bba
INFO [16:56:36.176] [bbotk] Evaluating 2 configuration(s)
INFO [16:56:36.185] [mlr3] Running benchmark with 20 resampling iterations
INFO [16:56:36.192] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/10)
INFO [16:56:36.221] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 2/10)
INFO [16:56:36.252] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 3/10)
INFO [16:56:36.285] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 4/10)
INFO [16:56:36.318] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 5/10)
INFO [16:56:36.347] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 6/10)
INFO [16:56:36.371] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 7/10)
INFO [16:56:36.396] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 8/10)
INFO [16:56:36.424] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 9/10)
INFO [16:56:36.448] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 10/10)
INFO [16:56:36.473] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/10)
INFO [16:56:36.500] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 2/10)
INFO [16:56:36.525] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 3/10)
INFO [16:56:36.552] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 4/10)
INFO [16:56:36.581] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 5/10)
INFO [16:56:36.601] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 6/10)
INFO [16:56:36.623] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 7/10)
INFO [16:56:36.645] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 8/10)
INFO [16:56:36.666] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 9/10)
INFO [16:56:36.689] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 10/10)
INFO [16:56:36.720] [mlr3] Finished benchmark
INFO [16:56:36.781] [bbotk] Result of batch 4:
INFO [16:56:36.783] [bbotk] cp minsplit classif.acc warnings errors runtime_learners
INFO [16:56:36.783] [bbotk] 0.05840653 10 0.7474710 0 0 0.13
INFO [16:56:36.783] [bbotk] 0.06122640 7 0.7448735 0 0 0.07
INFO [16:56:36.783] [bbotk] uhash
INFO [16:56:36.783] [bbotk] b3131398-4b84-4b0d-a2d4-cb6bd2759b25
INFO [16:56:36.783] [bbotk] 9d615792-9adc-4c26-95ac-ae04b22f8cd0
INFO [16:56:36.796] [bbotk] Evaluating 2 configuration(s)
INFO [16:56:36.803] [mlr3] Running benchmark with 20 resampling iterations
INFO [16:56:36.809] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/10)
INFO [16:56:36.828] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 2/10)
INFO [16:56:36.847] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 3/10)
INFO [16:56:36.866] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 4/10)
INFO [16:56:36.888] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 5/10)
INFO [16:56:36.909] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 6/10)
INFO [16:56:36.930] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 7/10)
INFO [16:56:36.952] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 8/10)
INFO [16:56:36.972] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 9/10)
INFO [16:56:36.995] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 10/10)
INFO [16:56:37.016] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/10)
INFO [16:56:37.038] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 2/10)
INFO [16:56:37.061] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 3/10)
INFO [16:56:37.085] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 4/10)
INFO [16:56:37.108] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 5/10)
INFO [16:56:37.131] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 6/10)
INFO [16:56:37.153] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 7/10)
INFO [16:56:37.177] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 8/10)
INFO [16:56:37.199] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 9/10)
INFO [16:56:37.218] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 10/10)
INFO [16:56:37.239] [mlr3] Finished benchmark
INFO [16:56:37.298] [bbotk] Result of batch 5:
INFO [16:56:37.300] [bbotk] cp minsplit classif.acc warnings errors runtime_learners
INFO [16:56:37.300] [bbotk] 0.02437254 9 0.7539645 0 0 0.10
INFO [16:56:37.300] [bbotk] 0.06133266 6 0.7448735 0 0 0.07
INFO [16:56:37.300] [bbotk] uhash
INFO [16:56:37.300] [bbotk] a000ae93-3cbb-48b2-9cef-747a4b315f40
INFO [16:56:37.300] [bbotk] 7f291572-7081-4da4-9a0b-cc45ffe576da
INFO [16:56:37.325] [bbotk] Finished optimizing after 10 evaluation(s)
INFO [16:56:37.326] [bbotk] Result:
INFO [16:56:37.328] [bbotk] cp minsplit learner_param_vals x_domain classif.acc
INFO [16:56:37.328] [bbotk] <num> <int> <list> <list> <num>
INFO [16:56:37.328] [bbotk] 0.02437254 9 <list[3]> <list[2]> 0.7539645auto_tuner调参的结果可以直接用于预测新数据
auto_learner$predict(task)<PredictionClassif> for 768 observations:
row_ids truth response
1 pos pos
2 neg neg
3 pos neg
--- --- ---
766 neg neg
767 pos neg
768 neg neg当我们拿到一份数据进行构建模型时，有很多信息对于目标变量来说，其实是无效的，如果把这些变量用于建模，只会增加噪声，降低模型的表现。
去除无效、冗余的变量，选择合适的变量的过程，就被称为特征选择
mlr3可按照一些方法，将所有预测变量计算一个分数，然后按照分数对变量进行排名和筛选
as.data.table(mlr_filters)Key: <key>
key label
<char> <char>
1: anova ANOVA F-Test
2: auc Area Under the ROC Curve Score
3: boruta Burota
4: carscore Correlation-Adjusted coRrelation Score
5: carsurvscore Correlation-Adjusted coRrelation Survival Score
6: cmim Minimal Conditional Mutual Information Maximization
7: correlation Correlation
8: disr Double Input Symmetrical Relevance
9: find_correlation Correlation-based Score
10: importance Importance Score
11: information_gain Information Gain
12: jmi Joint Mutual Information
13: jmim Minimal Joint Mutual Information Maximization
14: kruskal_test Kruskal-Wallis Test
15: mim Mutual Information Maximization
16: mrmr Minimum Redundancy Maximal Relevancy
17: njmim Minimal Normalised Joint Mutual Information Maximization
18: performance Predictive Performance
19: permutation Permutation Score
20: relief RELIEF
21: selected_features Embedded Feature Selection
22: univariate_cox Univariate Cox Survival Score
23: variance Variance
key label
task_types task_properties
<list> <list>
1: classif
2: classif twoclass
3: regr,classif
4: regr
5: surv
6: classif,regr
7: regr
8: classif,regr
9: NA
10: classif
11: classif,regr
12: classif,regr
13: classif,regr
14: classif
15: classif,regr
16: classif,regr
17: classif,regr
18: classif
19: classif
20: classif,regr
21: classif
22: surv
23: NA
task_types task_properties
params
<list>
1:
2:
3: pValue,mcAdj,maxRuns,doTrace,holdHistory,getImp,...
4: lambda,diagonal,verbose
5: maxIPCweight,denom
6: threads
7: use,method
8: threads
9: use,method
10: method
11: type,equal,discIntegers,threads
12: threads
13: threads
14: na.action
15: threads
16: threads
17: threads
18: method
19: standardize,nmc
20: neighboursCount,sampleSize
21: method
22:
23: na.rm
params
feature_types packages
<list> <list>
1: integer,numeric stats
2: integer,numeric mlr3measures
3: integer,numeric Boruta
4: logical,integer,numeric care
5: integer,numeric carSurv,mlr3proba
6: integer,numeric,factor,ordered praznik
7: integer,numeric stats
8: integer,numeric,factor,ordered praznik
9: integer,numeric stats
10: logical,integer,numeric,character,factor,ordered,... mlr3
11: integer,numeric,factor,ordered FSelectorRcpp
12: integer,numeric,factor,ordered praznik
13: integer,numeric,factor,ordered praznik
14: integer,numeric stats
15: integer,numeric,factor,ordered praznik
16: integer,numeric,factor,ordered praznik
17: integer,numeric,factor,ordered praznik
18: logical,integer,numeric,character,factor,ordered,... mlr3,mlr3measures
19: logical,integer,numeric,character,factor,ordered,... mlr3,mlr3measures
20: integer,numeric,factor,ordered FSelectorRcpp
21: logical,integer,numeric,character,factor,ordered,... mlr3
22: integer,numeric,logical survival
23: integer,numeric stats
feature_types packagesfilter <- flt("jmim")
task <- tsk("iris")
filter$calculate(task)
filter<FilterJMIM:jmim>: Minimal Joint Mutual Information Maximization
Task Types: classif, regr
Properties: -
Task Properties: -
Packages: praznik
Feature types: integer, numeric, factor, ordered
feature score
1: Petal.Width 1.0000000
2: Sepal.Length 0.6666667
3: Petal.Length 0.3333333
4: Sepal.Width 0.0000000task <- tsk("mtcars")
filter_cor <- flt("correlation")
filter_cor$param_set<ParamSet(2)>
id class lower upper nlevels default value
<char> <char> <num> <num> <num> <list> <list>
1: use ParamFct NA NA 5 everything [NULL]
2: method ParamFct NA NA 3 pearson [NULL]filter_cor$param_set$values <- list(method = "spearman")
filter_cor$param_set<ParamSet(2)>
id class lower upper nlevels default value
<char> <char> <num> <num> <num> <list> <list>
1: use ParamFct NA NA 5 everything [NULL]
2: method ParamFct NA NA 3 pearson spearmanfilter_cor$calculate(task)
filter_cor<FilterCorrelation:correlation>: Correlation
Task Types: regr
Properties: missings
Task Properties: -
Packages: stats
Feature types: integer, numeric
feature score
1: cyl 0.9108013
2: disp 0.9088824
3: hp 0.8946646
4: wt 0.8864220
5: vs 0.7065968
6: carb 0.6574976
7: drat 0.6514555
8: am 0.5620057
9: gear 0.5427816
10: qsec 0.4669358lrn <- lrn("classif.ranger", importance = "impurity")
task <- tsk("iris")
filter <- flt("importance", learner = lrn)
filter$calculate(task)
filter<FilterImportance:importance>: Importance Score
Task Types: classif
Properties: missings
Task Properties: -
Packages: mlr3, mlr3learners, ranger
Feature types: logical, integer, numeric, character, factor, ordered
feature score
1: Petal.Width 44.659090
2: Petal.Length 42.028596
3: Sepal.Length 9.992927
4: Sepal.Width 2.576583类似超参数调优，构建不同特征变量的模型，通过模型效果来选择
library(mlr3fselect)
task <- tsk("pima")
learner <- lrn("classif.rpart")
hout <- rsmp("holdout")
measure <- msr("classif.ce")
evals10 <- trm("evals", n_evals = 20)
instance <- mlr3fselect::FSelectInstanceBatchSingleCrit$new(
task = task,
learner = learner,
resampling = hout,
measure = measure,
terminator = evals10
)
instance<FSelectInstanceBatchSingleCrit>
* State: Not optimized
* Objective: <ObjectiveFSelectBatch:classif.rpart_on_pima>
* Terminator: <TerminatorEvals>fselector <- fs("random_search")
lgr::get_logger("bbotk")$set_threshold("warn")
fselector$optimize(instance)INFO [16:56:37.886] [mlr3] Running benchmark with 10 resampling iterations
INFO [16:56:37.892] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/1)
INFO [16:56:37.910] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/1)
INFO [16:56:37.926] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/1)
INFO [16:56:37.942] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/1)
INFO [16:56:37.959] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/1)
INFO [16:56:37.977] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/1)
INFO [16:56:37.992] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/1)
INFO [16:56:38.006] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/1)
INFO [16:56:38.027] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/1)
INFO [16:56:38.049] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/1)
INFO [16:56:38.066] [mlr3] Finished benchmark
INFO [16:56:38.238] [mlr3] Running benchmark with 10 resampling iterations
INFO [16:56:38.243] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/1)
INFO [16:56:38.262] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/1)
INFO [16:56:38.279] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/1)
INFO [16:56:38.296] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/1)
INFO [16:56:38.315] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/1)
INFO [16:56:38.333] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/1)
INFO [16:56:38.349] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/1)
INFO [16:56:38.376] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/1)
INFO [16:56:38.391] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/1)
INFO [16:56:38.408] [mlr3] Applying learner 'classif.rpart' on task 'pima' (iter 1/1)
INFO [16:56:38.424] [mlr3] Finished benchmark age glucose insulin mass pedigree pregnant pressure triceps
<lgcl> <lgcl> <lgcl> <lgcl> <lgcl> <lgcl> <lgcl> <lgcl>
1: TRUE TRUE TRUE TRUE TRUE TRUE TRUE TRUE
features n_features classif.ce
<list> <int> <num>
1: age,glucose,insulin,mass,pedigree,pregnant,... 8 0.2304688# 查看选中特征
instance$result_feature_set[1] "age" "glucose" "insulin" "mass" "pedigree" "pregnant" "pressure"
[8] "triceps" # 查看评估
instance$result_yclassif.ce
0.2304688 # 查看过程
as.data.table(instance$archive) age glucose insulin mass pedigree pregnant pressure triceps classif.ce
<lgcl> <lgcl> <lgcl> <lgcl> <lgcl> <lgcl> <lgcl> <lgcl> <num>
1: FALSE TRUE TRUE FALSE FALSE FALSE FALSE FALSE 0.2812500
2: TRUE TRUE TRUE TRUE TRUE FALSE FALSE FALSE 0.2617188
3: FALSE FALSE FALSE FALSE FALSE TRUE FALSE FALSE 0.3085938
4: TRUE TRUE TRUE TRUE TRUE TRUE TRUE TRUE 0.2304688
5: TRUE FALSE FALSE FALSE FALSE FALSE TRUE TRUE 0.3203125
6: TRUE FALSE FALSE FALSE TRUE TRUE FALSE FALSE 0.3203125
7: TRUE FALSE FALSE TRUE FALSE FALSE FALSE FALSE 0.3164062
8: FALSE FALSE TRUE FALSE FALSE FALSE FALSE FALSE 0.3945312
9: FALSE TRUE TRUE TRUE TRUE TRUE TRUE FALSE 0.2695312
10: TRUE FALSE FALSE FALSE TRUE FALSE FALSE FALSE 0.3125000
11: TRUE TRUE TRUE TRUE TRUE TRUE TRUE TRUE 0.2304688
12: TRUE TRUE TRUE TRUE TRUE TRUE TRUE TRUE 0.2304688
13: TRUE FALSE TRUE TRUE TRUE FALSE TRUE TRUE 0.3125000
14: FALSE TRUE TRUE FALSE TRUE TRUE FALSE FALSE 0.2656250
15: FALSE FALSE TRUE FALSE FALSE TRUE FALSE FALSE 0.3242188
16: TRUE TRUE TRUE TRUE TRUE TRUE TRUE TRUE 0.2304688
17: FALSE FALSE FALSE TRUE TRUE TRUE FALSE FALSE 0.3242188
18: TRUE FALSE TRUE FALSE TRUE FALSE TRUE FALSE 0.2617188
19: TRUE TRUE FALSE FALSE TRUE TRUE FALSE FALSE 0.2539062
20: TRUE FALSE FALSE TRUE FALSE TRUE FALSE TRUE 0.2968750
age glucose insulin mass pedigree pregnant pressure triceps classif.ce
runtime_learners timestamp batch_nr warnings errors
<num> <POSc> <int> <int> <int>
1: 0.00 2025-04-12 16:56:38 1 0 0
2: 0.02 2025-04-12 16:56:38 1 0 0
3: 0.01 2025-04-12 16:56:38 1 0 0
4: 0.02 2025-04-12 16:56:38 1 0 0
5: 0.01 2025-04-12 16:56:38 1 0 0
6: 0.00 2025-04-12 16:56:38 1 0 0
7: 0.00 2025-04-12 16:56:38 1 0 0
8: 0.00 2025-04-12 16:56:38 1 0 0
9: 0.01 2025-04-12 16:56:38 1 0 0
10: 0.02 2025-04-12 16:56:38 1 0 0
11: 0.00 2025-04-12 16:56:38 2 0 0
12: 0.00 2025-04-12 16:56:38 2 0 0
13: 0.00 2025-04-12 16:56:38 2 0 0
14: 0.02 2025-04-12 16:56:38 2 0 0
15: 0.01 2025-04-12 16:56:38 2 0 0
16: 0.02 2025-04-12 16:56:38 2 0 0
17: 0.02 2025-04-12 16:56:38 2 0 0
18: 0.00 2025-04-12 16:56:38 2 0 0
19: 0.01 2025-04-12 16:56:38 2 0 0
20: 0.02 2025-04-12 16:56:38 2 0 0
runtime_learners timestamp batch_nr warnings errors
features n_features resample_result
<list> <list> <list>
1: glucose,insulin 2 <ResampleResult>
2: age,glucose,insulin,mass,pedigree 5 <ResampleResult>
3: pregnant 1 <ResampleResult>
4: age,glucose,insulin,mass,pedigree,pregnant,... 8 <ResampleResult>
5: age,pressure,triceps 3 <ResampleResult>
6: age,pedigree,pregnant 3 <ResampleResult>
7: age,mass 2 <ResampleResult>
8: insulin 1 <ResampleResult>
9: glucose,insulin,mass,pedigree,pregnant,pressure 6 <ResampleResult>
10: age,pedigree 2 <ResampleResult>
11: age,glucose,insulin,mass,pedigree,pregnant,... 8 <ResampleResult>
12: age,glucose,insulin,mass,pedigree,pregnant,... 8 <ResampleResult>
13: age,insulin,mass,pedigree,pressure,triceps 6 <ResampleResult>
14: glucose,insulin,pedigree,pregnant 4 <ResampleResult>
15: insulin,pregnant 2 <ResampleResult>
16: age,glucose,insulin,mass,pedigree,pregnant,... 8 <ResampleResult>
17: mass,pedigree,pregnant 3 <ResampleResult>
18: age,insulin,pedigree,pressure 4 <ResampleResult>
19: age,glucose,pedigree,pregnant 4 <ResampleResult>
20: age,mass,pregnant,triceps 4 <ResampleResult>
features n_features resample_result# 将选中变量应用于模型
task$select(instance$result_feature_set) # 只使用选中的变量
learner$train(task)task <- tsk("penguins")
split <- partition(task, ratio = 0.8)
afs <- auto_fselector(
fselector = fs("random_search"),
learner = lrn("classif.rpart"),
resampling = rsmp("holdout"),
measure = msr("classif.ce"),
term_evals = 4
)
afs$train(task, row_ids = split$train)INFO [16:56:38.759] [mlr3] Running benchmark with 10 resampling iterations
INFO [16:56:38.764] [mlr3] Applying learner 'classif.rpart' on task 'penguins' (iter 1/1)
INFO [16:56:38.777] [mlr3] Applying learner 'classif.rpart' on task 'penguins' (iter 1/1)
INFO [16:56:38.795] [mlr3] Applying learner 'classif.rpart' on task 'penguins' (iter 1/1)
INFO [16:56:38.810] [mlr3] Applying learner 'classif.rpart' on task 'penguins' (iter 1/1)
INFO [16:56:38.827] [mlr3] Applying learner 'classif.rpart' on task 'penguins' (iter 1/1)
INFO [16:56:38.844] [mlr3] Applying learner 'classif.rpart' on task 'penguins' (iter 1/1)
INFO [16:56:38.862] [mlr3] Applying learner 'classif.rpart' on task 'penguins' (iter 1/1)
INFO [16:56:38.879] [mlr3] Applying learner 'classif.rpart' on task 'penguins' (iter 1/1)
INFO [16:56:38.895] [mlr3] Applying learner 'classif.rpart' on task 'penguins' (iter 1/1)
INFO [16:56:38.913] [mlr3] Applying learner 'classif.rpart' on task 'penguins' (iter 1/1)
INFO [16:56:38.930] [mlr3] Finished benchmarkafs$predict(task, row_ids = split$test)<PredictionClassif> for 69 observations:
row_ids truth response
9 Adelie Adelie
11 Adelie Adelie
18 Adelie Gentoo
--- --- ---
326 Chinstrap Chinstrap
336 Chinstrap Chinstrap
339 Chinstrap Chinstrap通过mlr3pipelines可以将数据预处理、建模、模型比较、集成学习结合在一起
将数据预处理、特征选择结合
graph <-
po("imputehist", # 插补
id = "impute_num", # 重命名步骤
affect_columns = is.numeric # 设置处理哪些列
) %>>%
po("imputeoor", id = "impute_fct", affect_columns = is.factor) %>>% # 插补因子
po("filter", mlr3filters::flt("information_gain"),
filter.frac = 0.95
) %>>%
po("encode", method = "one-hot") %>>%
po("learner", lrn("classif.rpart"))
graph$plot()
task <- tsk("pima")
lrn_graph <- as_learner(graph)
lrn_graph$train(task)
pred <- lrn_graph$predict(task)
pred$confusion truth
response pos neg
pos 207 73
neg 61 427针对数据量较大时，可以将数据分块训练，最后将各块
chks <- po("chunk", 4)
lrns <- ppl("greplicate", po("learner", lrn("classif.rpart")), 4)
mjv <- po("classifavg", 4)
pipeline <- chks %>>% lrns %>>% mjv
pipeline$plot(html = FALSE)
task <- tsk("iris")
split <- partition(task, ratio = 0.7)
pipelrn <- as_learner(pipeline)
pipelrn$train(task, split$train)$
predict(task, split$test)$
score(msr("classif.acc"))classif.acc
0.7333333 single_pred <- po("subsample", frac = 0.7) %>>%
po("learner", lrn("classif.rpart")) # 建立一个模型
pred_set <- ppl("greplicate", single_pred, 10L) # 复制10次
bagging <- pred_set %>>%
po("classifavg", innum = 10)
bagging$plot(html = FALSE)
task <- tsk("iris")
split <- partition(task, ratio = 0.7)
baglrn <- as_learner(bagging)
baglrn$train(task, row_ids = split$train)
baglrn$predict(task, row_ids = split$test)$
score(msr("classif.acc"))classif.acc
0.9777778 lrn <- lrn("classif.rpart")
lrn_0 <- po("learner_cv", lrn$clone())
lrn_0$id <- "rpart_cv"
level_0 <- gunion(list(lrn_0, po("nop")))
combined <- level_0 %>>% po("featureunion", 2)
stack <- combined %>>% po("learner", lrn$clone())
stack$plot(html = FALSE)
stacklrn <- as_learner(stack)
stacklrn$train(task, split$train)INFO [16:56:41.865] [mlr3] Applying learner 'classif.rpart' on task 'iris' (iter 1/3)
INFO [16:56:41.880] [mlr3] Applying learner 'classif.rpart' on task 'iris' (iter 2/3)
INFO [16:56:41.894] [mlr3] Applying learner 'classif.rpart' on task 'iris' (iter 3/3)stacklrn$predict(task, split$test)$
score(msr("classif.acc"))classif.acc
0.9555556 library("magrittr")
library("mlr3learners")
rprt <- lrn("classif.rpart", predict_type = "prob")
glmn <- lrn("classif.glmnet", predict_type = "prob")
# 创建学习器
lrn_0 <- po("learner_cv", rprt, id = "rpart_cv_1")
lrn_0$param_set$values$maxdepth <- 5L
lrn_1 <- po("pca", id = "pca1") %>>% po("learner_cv", rprt, id = "rpart_cv_2")
lrn_1$param_set$values$rpart_cv_2.maxdepth <- 1L
lrn_2 <- po("pca", id = "pca2") %>>% po("learner_cv", glmn)
# 第0层
level_0 <- gunion(list(lrn_0, lrn_1, lrn_2, po("nop", id = "NOP1")))
# 第1层
level_1 <- level_0 %>>%
po("featureunion", 4) %>>%
po("copy", 3) %>>%
gunion(list(
po("learner_cv", rprt, id = "rpart_cv_l1"),
po("learner_cv", glmn, id = "glmnt_cv_l1"),
po("nop", id = "NOP_l1")
))
# 第2层
level_2 <- level_1 %>>%
po("featureunion", 3, id = "u2") %>>%
po("learner", rprt, id = "rpart_l2")
level_2$plot(html = FALSE)
task <- tsk("iris")
lrn <- as_learner(level_2)
lrn$train(task, split$train)$
predict(task, split$test)$
score(msr("classif.acc"))INFO [16:56:42.579] [mlr3] Applying learner 'classif.rpart' on task 'iris' (iter 1/3)
INFO [16:56:42.595] [mlr3] Applying learner 'classif.rpart' on task 'iris' (iter 2/3)
INFO [16:56:42.613] [mlr3] Applying learner 'classif.rpart' on task 'iris' (iter 3/3)
INFO [16:56:42.715] [mlr3] Applying learner 'classif.rpart' on task 'iris' (iter 1/3)
INFO [16:56:42.742] [mlr3] Applying learner 'classif.rpart' on task 'iris' (iter 2/3)
INFO [16:56:42.768] [mlr3] Applying learner 'classif.rpart' on task 'iris' (iter 3/3)
INFO [16:56:43.059] [mlr3] Applying learner 'classif.glmnet' on task 'iris' (iter 1/3)
INFO [16:56:43.106] [mlr3] Applying learner 'classif.glmnet' on task 'iris' (iter 2/3)
INFO [16:56:43.142] [mlr3] Applying learner 'classif.glmnet' on task 'iris' (iter 3/3)INFO [16:56:43.253] [mlr3] Applying learner 'classif.rpart' on task 'iris' (iter 1/3)
INFO [16:56:43.291] [mlr3] Applying learner 'classif.rpart' on task 'iris' (iter 2/3)
INFO [16:56:43.335] [mlr3] Applying learner 'classif.rpart' on task 'iris' (iter 3/3)
INFO [16:56:43.461] [mlr3] Applying learner 'classif.glmnet' on task 'iris' (iter 1/3)
INFO [16:56:43.524] [mlr3] Applying learner 'classif.glmnet' on task 'iris' (iter 2/3)
INFO [16:56:43.588] [mlr3] Applying learner 'classif.glmnet' on task 'iris' (iter 3/3)classif.acc
0.9555556