Fukuoka.R#15 https://fukuoka-r.connpass.com/event/139211/ での発表資料です リポジトリ http://github.com/uribo/190831_fukuokar15
Uryu Shinya @u_ribotidymodelsʹΑΔϞσϧߏஙͱӡ༻190831 Fukuoka.R@LINEԬ
View Slide
࣍UJEZNPEFMTʹΑΔϞσϧߏஙϞσϧͷվળͱӡ༻σʔλϞσϦϯάͷϫʔΫϑϩʔ123
σʔλϞσϦϯάͷϫʔΫϑϩʔtidymodelsʹΑΔϞσϧߏஙͱӡ༻190831 Fukuoka.R@LINEԬ
Garrett and Hadley (2016)ΑΓ࡞tidymodels͜͜σʔλੳͷϫʔΫϑϩʔ
෮తͳ࡞ۀͰϞσϧΛຏ্͖͍͛ͯ͘ϞσϧߏஙͷయܕతͳΈMax and Kjell (2019)ΑΓ࡞
B σʔλͷಛɺσʔλؒͷؔΛΓɺॳظϞσϧʹར༻͢Δʮग़ൃʯΛݟ͚ͭΔͨΊͷࢹ֮ԽɻC ౷ܭྔͷूܭతมͱڧ͍૬ؔͷ͋ΔมΛಛఆɺϞσϧʹର͢ΔԾઆΛཱͯΔɻσʔλΛेʹཧղͰ͖ͨͱݴ͑Δ·ͰɺؔΛՄࢹԽ͠ɺ͞ΒͳΔఆྔੳΛ܁Γฦ͢ɻD σʔλΛॳظϞσϧʹద༻͢ΔͨΊͷ४උɻ͍͟ɺϞσϦϯάͷ࣮ߦʂͱ͍͔ͳ͍
E ॳظϞσϧͷ࣮ߦɻॳظϞσϧʹར༻ͨ͠σʔλͰɺɹɹɹ͍͔ͭ͘ͷϞσϧద༻ɺൺֱɻɹɹɹɹɹɹɹɹɹɹɹɹϋΠύʔύϥϝʔλͷ୳ࡧ͜͜ͰߦΘΕΔɻF ෳճߦΘΕͨύϥϝʔλௐͷ݁ՌΛੳG Ϟσϧͷ݁ՌΛՄࢹԽෳͷϞσϧͰͷੑೳΛൺֱ͢Δ
H ॳظϞσϧΛվྑ͢ΔಛྔΤϯδχΞϦϯάI ࠷ऴతͳީิϞσϧʹର͢ΔௐJ ධՁηοτΛར༻ͨ͠൚ԽੑೳͷධՁK ӡ༻ϞσϧΛվળ͢ΔಛྔΛ୳͢
ϞσϧͷੑೳධՁλεΫʹԠͨ͡ධՁࢦඪΛར༻͢Δܾఆ(R2, RSQ: coefficient of determination)ೋฏۉฏํࠜޡࠩ(RMSE: Root Mean Square Error)ฏۉઈରޡࠩ(MAE: Mean absolute error)ࠞಉߦྻਖ਼ղద߹ͱ࠶ݱ ROCۂઢͱAUCճؼྨ
ৄͪ͘͜͠Βhttp://bit.ly/slide-fe-recipes\TweetStar͕ྭΈʹͳΓ·͢ʂ/http://bit.ly/practical-ds
ͰσʔλϞσϦϯάΛߦ͏ࡍͷ՝ଟ͘ͷύοέʔδ͕։ൃ͞Ε͍ͯΔ͕ΠϯλʔϑΣΠεʹ౷Ұੑ͕ͳ͍ϞσϧΦϒδΣΫτΛѻ͏ͨΊͷformulaͷܽ5IF3'PSNVMB.FUIPE5IF(PPE1BSUTu37JFXTIUUQTSWJFXTSTUVEJPDPNUIFSGPSNVMBNFUIPEUIFHPPEQBSUT5IF3'PSNVMB.FUIPE5IF#BE1BSUTu37JFXTIUUQTSWJFXTSTUVEJPDPNUIFSGPSNVMBNFUIPEUIFCBEQBSUT
tidymodelsͷ֓ཁtidymodelsʹΑΔϞσϧߏஙͱӡ༻190831 Fukuoka.R@LINEԬ
UJEZNPEFMT.BY,VIO 34UVEJP\DBSFU^ͷ։ൃऀΒ͕ϝϯόʔUJEZͳ౷ܭϞσϦϯάػցֶशͷॲཧΛߦ͏ύοέʔδ܈Λఏڙ͢ΔύΠϓϑϨϯυϦʔਓؒʹϓϩάϥϜʹ༏͍͠ઃܭ
{parsnip}{recipes}{rsample}{yardstick}{tidymodels}ʹؚ·ΕΔύοέʔδlibrary(tidymodels)#> Registered S3 method overwritten by 'xts':#> method from#> as.zoo.xts zoo#> ─ Attaching packages ───── tidymodels 0.0.2 ─#> ✔ broom 0.5.2 ✔ purrr 0.3.2#> ✔ dials 0.0.2 ✔ recipes 0.1.6#> ✔ dplyr 0.8.3 ✔ rsample 0.0.5#> ✔ ggplot2 3.2.1 ✔ tibble 2.1.3#> ✔ infer 0.4.0.1 ✔ yardstick 0.0.3#> ✔ parsnip 0.0.3.1#> ─ Conflicts ──────── tidymodels_conflicts() ─#> ✖ purrr ::discard() masks scales ::discard()#> ✖ dplyr ::filter() masks stats ::filter()#> ✖ dplyr ::lag() masks stats ::lag()#> ✖ recipes ::step() masks stats ::step()Ϟσϧߏஙɾద༻ϞσϧͷੑೳධՁσʔλલॲཧɺಛྔੜϦαϯϓϦϯάಡΈࠐ·ΕΔύοέʔδɺόʔδϣϯ͓Αͼ໊ؔͷিಥʹ͍ͭͯग़ྗ
ΞϠϝଐछͷྨछ໊(Species)Λϥϕϧͱ͠ɺྨΛద༻͢Δiris <- as_tibble(iris)glimpse(iris)#> Observations: 150#> Variables: 5#> $ Sepal.Length 5.1, 4.9, 4.7, 4.6, 5.0, 5.4, 4.6, …#> $ Sepal.Width 3.5, 3.0, 3.2, 3.1, 3.6, 3.9, 3.4, …#> $ Petal.Length 1.4, 1.4, 1.3, 1.5, 1.4, 1.7, 1.4, …#> $ Petal.Width 0.2, 0.2, 0.2, 0.2, 0.2, 0.4, 0.3, …#> $ Species setosa, setosa, setosa, setosa, se…Ֆͷܗଶʹؔ͢Δͭͷಛྔ
σʔληοτΛੳηοτ USBJOɺධՁηοτ UFTUʹ͚ΔੳηοτධՁηοτσʔλׂ3ͰͷφͳσʔλׂͷΓํSTBNQMFύοέʔδʹΑΔަࠩݕূגࣜձࣾϗΫιΤϜͷϒϩάIUUQTCMPHIPYPNDPNFOUSZ{rsample}σʔληοτϞσϧͷֶशʹ༻͍Δ ϞσϧͷੑೳධՁΛଌఆ͢ΔͨΊɺະͷใͱͯ͠ར༻͞ΕΔ
ׂϥϯμϜσʔλׂ {rsample}σʔληοτͷׂΛੳηοτͱ͢Δiris_split <- initial_split(iris, prop = 0.6)iris_split#> <90/60/150>iris_train <- training(iris_split)iris_test <- testing(iris_split)
લॲཧɾಛྔΤϯδχΞϦϯά{recipes}Ϟσϧʹ༻͍ΔσʔλՃͷखଓ͖ΛʮϨγϐʯԽϞσϧͰѻ͏σʔλͷલॲཧΛSFDJQFTͰߦ͏גࣜձࣾϗΫιΤϜͷϒϩάIUUQTCMPHIPYPNDPNFOUSZ123recipe()step_*()prep()bake()/juice()4ར༻͢ΔมͷؔΛఆٛˠࡐྉΛࢦఆ͢ΔσʔλՃͷखଓ͖Λࢦఆˠௐཧ๏Λهड़͢Δσʔληοτʹద༻ˠௐཧΛߦ͏TUFQ@ ͷॲཧΛ౷߹ˠϨγϐΛ֬ೝ͢Δ
એʢٳܜʣλΠϜhttps://uribo.github.io/dpp-cookbook/
iris_recipe <-iris_train %>%#> #4QFDJFTΛతมɺଞͷมΛઆ໌มʹͨ͠Ϟσϧrecipe(formula = Species ~ .) %>%#> # εςοϓ1: ͯ͢ͷઆ໌มΛରʹ#> # ૬ؔͷߴ͍(0.9Ҏ্)มͷ͍ͣΕ͔Λআ֎step_corr(all_predictors(), threshold = 0.9) %>%#> # εςοϓ2: ͯ͢ͷઆ໌มΛରʹมͷඪ४Խstep_normalize(all_predictors(), -all_outcomes())ϞσϧͷॲཧΛύΠϓԋࢉࢠͰՃoutcome ~ predictorsrole
step_*()ͰͷมͷࢦఆจࣈྻͰͷࢦఆtidyselectͷؔϞσϧͰͷrole123มͷσʔλܕ4all_predictors() all_outcomes()starts_with() contains()ͳͲall_nominal() all_numeric()"Species" "Sepal.Length"
recipeΦϒδΣΫτiris_recipe#> Data Recipe#>#> Inputs:#>#> role #variables#> outcome 1#> predictor 4#>#> Operations:#>#> Correlation filter on all_predictors#> Centering and scaling for all_predictors, -, all_outcomes()recipe()͓Αͼstep_*()Ͱఆٛͨ͠खॱΛࣔ͢
iris_recipe_comp <-iris_recipe %>%prep()#> … தུ#> Training data contained 90 data points and no missing data.#>#> Operations:#>#> Correlation filter removed Petal.Length [trained]#> Centering and scaling for Sepal.Length, Sepal.Width,Petal.Width [trained]ੳηοτʹద༻ͨ͠ΒͲ͏ͳΔ͔͜ͷঢ়ଶͰ·ͩσʔλ͕ੜ͞Ε͍ͯͳ͍recipeΦϒδΣΫτ
σʔληοτʹϨγϐΛద༻iris_training <-juice(iris_recipe_comp)glimpse(iris_training)#> Observations: 90#> Variables: 4#> $ Sepal.Length -1.35774587, -1.71228299…#> $ Sepal.Width 0.230356130, -0.438419732…#> $ Petal.Width -1.2122997, -1.2122997…#> $ Species setosa, se…iris_testing <-iris_recipe_comp %>%bake(new_data = iris_test)ੳηοτ ධՁηοτPetal.Length͕ফ͍͑ͯΔiris_testing Ͱಉ༷
all.equal(mean(scale(iris_train$Sepal.Length)),mean(iris_training$Sepal.Length))#> [1] TRUE{recipes}ͷॲཧ༰Λ͔֬ΊΔฏۉ ࢄͷඪ४Խiris_train %>%recipe(formula = Species ~ .) %>%step_corr(all_predictors(), threshold = 0.9) %>%step_normalize(all_predictors(), -all_outcomes())
{recipes}ͷॲཧ༰Λ͔֬ΊΔcorrr ::correlate(iris_test %>%select_if(is.numeric),method = "pearson",use = "pairwise.complete.obs",quiet = TRUE)#> # A tibble: 4 x 5#> rowname Sepal.Length Sepal.Width Petal.Length Petal.Width#> #> 1 Sepal.Length NA -0.0685 0.859 0.798#> 2 Sepal.Width -0.0685 NA -0.395 -0.333#> 3 Petal.Length 0.859 -0.395 NA 0.958#> 4 Petal.Width 0.798 -0.333 0.958 NA૬ؔͷߴ͍มͷยํΛσʔληοτ͔Βআ֎͢Δ
recipes͕ఏڙ͢ΔstepมΤϯίʔσΟϯάɾ࣌ؒϑΟϧλॲཧstep_*()ؔͱͯ͠ఏڙ͞ΕΔܽଛͷೖඪ४Խ࣍ݩݮͳͲls("package:recipes", pattern = “^step_")#> # 69ݸͷstep_*ؔ (version 0.1.6)
step_*()ͷ֦ுͱ࣮σʔλܕʹԠͨ͡ {textrecipes}{embed}5FOTPS'MPXϞσϧͰར༻ {tfdatasets}step_*()Λఏڙ͢ΔύοέʔδจࣈྻΧςΰϦຊޠͷલॲཧ{washoku}ਖ਼نԽɺ֯શ֯ॲཧɺॅॴׂͳͲߏதremotes ::install_github("uribo/washoku")
Ϟσϧߏங {parsnip}ଟ༷ͳϞσϦϯάύοέʔδΛϥοϓɻ౷Ұతʹૢ࡞Մೳ༷ΛఆٛΤϯδϯʢύοέʔδʣΛࢦఆϞσϧͷͯΊճؼʁྨʁ1set_engine()՝ʹదͨ͠ϞσϧΛબͿ23 fit()rand_forest() linear_reg()logistic_reg()ੳηοτͷֶश
ύοέʔδܾఆʹ༩͑Δಛྔ࡞͢Δܾఆͷϊʔυʹؚ·ΕΔ࠷খͷαϯϓϧranger mtry num.trees min.node.sizerandomForest mtry ntree nodesizesparklyr mtry num.trees min_instances_per_nodeؔͷॻ͖ํʢҾͷࢦఆํ๏ʣ͕ύοέʔδʹΑͬͯҟͳΔ{parsnip}ΛΘͳ͍߹rand_forest(mode = "classification", mtry, trees, min_n)
rf_spec <-rand_forest(mode = "classification", trees = 100)#> #ranger ::rangeriris_ranger <-rf_spec %>%set_engine("ranger", seed = 123) %>%fit(Species ~ ., data = iris_training)#> #randomForest ::randomForestiris_rf <-rf_spec %>%set_engine("randomForest") %>%fit(Species ~ ., data = iris_training)ϥϯμϜϑΥϨετΛར༻ͨ͠ϞσϧΤϯδϯ ύοέʔδʹݻ༗ͷΦϓγϣϯΛࢦఆՄೳ
Ϟσϧͷద༻݁Ռiris_rf#> Call:#> randomForest(x = as.data.frame(x), y = y, ntree = ~100)#> Type of random forest: classification#> Number of trees: 100#> No. of variables tried at each split: 1#>#> OOB estimate of error rate: 2.22%#> Confusion matrix:#> setosa versicolor virginica class.error#> setosa 36 0 0 0.00000000#> versicolor 0 26 1 0.03703704#> virginica 0 1 26 0.03703704rand_forest()ɺset_engine()Ͱࢦఆͨ͠ॲཧ͕࣮ߦ
Ϟσϧͷվળͱӡ༻tidymodelsʹΑΔϞσϧߏஙͱӡ༻190831 Fukuoka.R@LINEԬ
{yardstick}ϞσϧੑೳධՁςετσʔλʢ৽͍͠σʔλʣΛରʹϥϕϧͷਪఆiris_rf_pred <-predict(iris_rf, iris_testing) %>%bind_cols(iris_testing)rf_metrics <-iris_rf_pred %>%metrics(truth = Species, estimate = .pred_class)rf_metrics#> # A tibble: 2 x 3#> .metric .estimator .estimate#> #> 1 accuracy multiclass 0.95#> 2 kap multiclass 0.925ྨͷλεΫͰਫ਼ BDDVSBDZͱ,BQQB͕ੑೳࢦඪͱͯ͠༻ҙ͞ΕΔ
{yardstick}ੑೳࢦඪͷΈ߹ΘͤͨΓಛఆͷੑೳࢦඪΛબΔϞσϧੑೳධՁྨͷλεΫͰͷੑೳධՁؔʹUSVUI FTUJNBUFΛࢦఆ͢Δiris_rf_pred %>%accuracy(Species, .pred_class)perf_metrics <- metric_set(accuracy, recall, precision)iris_rf_pred %>%perf_metrics(truth = Species, estimate = .pred_class)#> # A tibble: 3 x 3#> .metric .estimator .estimate#> #> 1 accuracy multiclass 0.983#> 2 recall macro 0.981#> 3 precision macro 0.986iris_rf_pred %>%kap(Species, .pred_class)
ggplotϕʔεͷՄࢹԽautoplot(iris_rf,type = "heatmap")ࠞಉߦྻͷώʔτϚοϓ
ggplotϕʔεͷՄࢹԽiris_ranger %>%predict(iris_testing, type = "prob") %>%bind_cols(iris_testing) %>%roc_curve(Species, rstarts_with(".pred")) %>%autoplot()ROCۂઢ
σʔλͷׂҰ͖ΓަࠩݕূͷͨΊͷϦαϯϓϦϯά {rsample}ੳηοτධՁηοτiris_split <- initial_split(iris, prop = 0.6)ෳͷϦαϯϓϦϯάσʔλ͔ΒϞσϧͷ൚ԽੑೳΛධՁ͢Δ͜ͱ͕ඞཁˠަࠩݕূ
σʔλɺతʹԠׂͨ͡ํ๏ΛબͿ࣌ܥྻɺۭؒσʔλͳͲҙ
iris_cv <- vfold_cv(iris, v = 4)iris_cv#> # A tibble: 4 x 2#> splits id#> #> 1 Fold1#> 2 Fold2#> 3 Fold3#> 4 Fold4σʔλɺతʹԠׂͨ͡ํ๏ΛબͿassessment(iris_cv$splits[[1]])ධՁηοτanalysis(iris_cv$splits[[1]])ੳηοτFoldʹϦαϯϓϦϯά͞ΕͨੳηοτɺධՁηοτΛࢀরσʔλϑϨʔϜͷߦ͝ͱʹׂ͞Εͨσʔληοτؚ͕·ΕΔ
Fold͝ͱʹֶशɺੑೳࢦඪΛࢉग़#> # 4-fold cross-validation#> # A tibble: 4 x 4#> splits id recipes accuracy#> * #> 1 Fold1 0.974#> 2 Fold2 0.895#> 3 Fold3 0.919#> 4 Fold4 0.973.Last.value %>%pull(accuracy) %>%mean()#> [1] 0.9400782
ϥϯμϜϑΥϨετϞσϧͰͭͷϋΠύʔύϥϝʔλͷࢦఆ͕ՄೳϋΠύʔύϥϝʔλͷ୳ࡧ {dials}ϋΠύʔύϥϝʔλͷ͕Ϟσϧͷਫ਼ʹӨڹ͢Δrand_forest(mode = "classification", mtry, trees, min_n)ύϥϝʔλͷΛมԽͤͨ͞ঢ়ଶͰͷੑೳධՁ͕ඞཁ
set.seed(1234)bst_grid <- grid_random(range_set(trees, c(1, 50)),range_set(min_n, c(2, 30)),size = 3)#> # A tibble: 3 x 2#> trees min_n#> #> 1 28 6#> 2 16 13#> 3 22 16άϦουαʔνmod_obj <-rand_forest(ɹɹmode = "classification",ɹɹtrees = varying(),ɹɹmin_n = varying(),ɹɹmtry = 3)ύϥϝʔλ໊ͱϥϯμϜʹׂΓৼΔͷൣғΛࢦఆTJ[F݅ͷϥϯμϜσʔλ͕ੜ{parsnip} ͰͷϞσϧ࡞ஈ֊Ͱݻఆ͠ͳ͍ύϥϝʔλΛࢦఆ
lapply(), for()ͰOKͰ͢purrrͱΈ߹Θͤͯseq.int(nrow(bst_grid)) %>%purrr ::map(~ merge(mod_obj, bst_grid[.x, ])) %>%purrr ::flatten()#>#> Random Forest Model Specification (classification)#>#> Main Arguments:#> mtry = 3#> trees = 28#> min_n = 6#> Main Arguments:#> mtry = 3#> trees = 16#> min_n = 13#> Main Arguments:#> mtry = 3#> trees = 22#> min_n = 16ͦΖͦΖखΛग़͢QVSSSOFLPTLZ4QFBLFS%FDLIUUQTTQFBLFSEFDLDPNT@VSZVOFLPTLZ
·ͱΊ{parsnip}{recipes}{rsample} {yardstick}Ϟσϧͷֶश ੑೳධՁσʔλՃϦαϯϓϦϯά{tidymodels} ύΠϓϑϨϯυϦʔɺ౷ܭϞσϧɾػցֶशʹඞཁͳॲཧΛ؆ུԽͰ͖Δ౷ҰతͳΠϯλʔϑΣʔεΛఏڙ͢Δɻinitial_split()vfold_cv()step_*()bake()/juice()set_engine() metrics()1 2 3 4
&OKPZָ͠ΜͰ͘ΕΜ
s_uryu
ohmae
lnit
loic425
ari_mharada
psj59129
murabayashi
sr2mg4
shimabox
ogiwarat
gmvalentino8
mrasu
tosite
paulrobertlloyd
wjessup
zakiyama
ammeep
lemiorhan
addyosmani
3n
thekraken
sachag
tenderlove
denniskardys