[英]Tuning with classification_cost and custom cost matrix in Tidymodels
我正在使用 tidymodels 來構建一個模型,其中假陰性比假陽性成本更高。 因此,我想使用yardstick::classification_cost
指標進行超參數調整,但使用反映這一事實的自定義分類成本矩陣。
在擬合模型后執行此操作非常簡單:
library(tidymodels)
# load simulated prediction output
data("two_class_example")
# cost matrix penalizing false negatives
cost_matrix <- tribble(
~truth, ~estimate, ~cost,
"Class1", "Class2", 2,
"Class2", "Class1", 1
)
# use function on simulated prediction output
classification_cost(
data = two_class_example,
truth = truth,
# target class probability
Class1,
# supply the function with the cost matrix
costs = cost_matrix)
#> # A tibble: 1 × 3
#> .metric .estimator .estimate
#> <chr> <chr> <dbl>
#> 1 classification_cost binary 0.260
由reprex 包(v2.0.1) 於 2021 年 11 月 1 日創建
但是在超參數調整期間使用這個函數是我遇到問題的地方。 該文檔指出,對於設置選項,指標應包含在自定義函數中。 這是我的嘗試和由此產生的錯誤。 請注意此包裝器如何在評估擬合模型時正常工作,但在嘗試用於調整時會引發錯誤:
library(tidymodels)
# load data
data("two_class_example")
data("two_class_dat")
# create custom metric penalizing false negatives
classification_cost_penalized <- function(
data,
truth,
class_proba,
na_rm = TRUE
) {
# cost matrix penalizing false negatives
cost_matrix <- tribble(
~truth, ~estimate, ~cost,
"Class1", "Class2", 2,
"Class2", "Class1", 1
)
classification_cost(
data = data,
truth = !! rlang::enquo(truth),
# supply the function with the class probabilities
!! rlang::enquo(class_proba),
# supply the function with the cost matrix
costs = cost_matrix,
na_rm = na_rm
)
}
# Use `new_numeric_metric()` to formalize this new metric function
classification_cost_penalized <- new_prob_metric(classification_cost_penalized, "minimize")
# test if this works on the simulated estimates
two_class_example %>%
classification_cost_penalized(truth = truth, class_prob = Class1)
#> # A tibble: 1 × 3
#> .metric .estimator .estimate
#> <chr> <chr> <dbl>
#> 1 classification_cost binary 0.260
# test if this works with hyperparameter tuning
# specify a RF model
my_model <-
rand_forest(mtry = tune(),
min_n = tune(),
trees = 500) %>%
set_engine("ranger") %>%
set_mode("classification")
# specify recipe
my_recipe <- recipe(Class ~ A + B, data = two_class_dat)
# bundle to workflow
my_wf <- workflow() %>%
add_model(my_model) %>%
add_recipe(my_recipe)
# start tuning
tuned_rf <- my_wf %>%
# set up tuning grid
tune_grid(
resamples = vfold_cv(two_class_dat,
v = 5),
grid = 5,
metrics = metric_set(classification_cost_penalized))
#> i Creating pre-processing data to finalize unknown parameter: mtry
#> x Fold1: internal: Error: In metric: `classification_cost_penalized`
#> unused argum...
#> x Fold2: internal: Error: In metric: `classification_cost_penalized`
#> unused argum...
#> x Fold3: internal: Error: In metric: `classification_cost_penalized`
#> unused argum...
#> x Fold4: internal: Error: In metric: `classification_cost_penalized`
#> unused argum...
#> x Fold5: internal: Error: In metric: `classification_cost_penalized`
#> unused argum...
#> Warning: All models failed. See the `.notes` column.
由reprex 包(v2.0.1) 於 2021 年 11 月 1 日創建
解開注釋顯示有未使用的參數: "internal: Error: In metric:
classification_cost_penalized \\nunused arguments (estimator = ~prob_estimator, event_level = ~event_level)"
但顯然\\nunused arguments (estimator = ~prob_estimator, event_level = ~event_level)"
yardstick_event_level()
函數,這就是event_level
應該如何設置到這個文檔,不存在? 搜索時沒有顯示該名稱下的功能。
我不知道如何在這里進行。 感謝您的時間。
當您調整現有的衡量指標時,使用metric_tweak()
函數要容易得多,它允許您對某些可選參數(如cost
)進行硬編碼,同時保持其他所有參數相同。 它有點像purrr::partial()
,但用於衡量指標。
library(tidymodels)
# load data
data("two_class_example")
data("two_class_dat")
cost_matrix <- tribble(
~truth, ~estimate, ~cost,
"Class1", "Class2", 2,
"Class2", "Class1", 1
)
classification_cost_penalized <- metric_tweak(
.name = "classification_cost_penalized",
.fn = classification_cost,
costs = cost_matrix
)
# test if this works on the simulated estimates
two_class_example %>%
classification_cost_penalized(truth = truth, class_prob = Class1)
#> # A tibble: 1 × 3
#> .metric .estimator .estimate
#> <chr> <chr> <dbl>
#> 1 classification_cost_penalized binary 0.260
# specify a RF model
my_model <-
rand_forest(
mtry = tune(),
min_n = tune(),
trees = 500
) %>%
set_engine("ranger") %>%
set_mode("classification")
# specify recipe
my_recipe <- recipe(Class ~ A + B, data = two_class_dat)
# bundle to workflow
my_wf <- workflow() %>%
add_model(my_model) %>%
add_recipe(my_recipe)
# start tuning
tuned_rf <- my_wf %>%
tune_grid(
resamples = vfold_cv(two_class_dat, v = 5),
grid = 5,
metrics = metric_set(classification_cost_penalized)
)
#> i Creating pre-processing data to finalize unknown parameter: mtry
collect_metrics(tuned_rf)
#> # A tibble: 5 × 8
#> mtry min_n .metric .estimator mean n std_err .config
#> <int> <int> <chr> <chr> <dbl> <int> <dbl> <chr>
#> 1 1 35 classification_cost… binary 0.407 5 0.0162 Preprocessor1…
#> 2 1 23 classification_cost… binary 0.403 5 0.0146 Preprocessor1…
#> 3 1 10 classification_cost… binary 0.403 5 0.0137 Preprocessor1…
#> 4 2 27 classification_cost… binary 0.396 5 0.0166 Preprocessor1…
#> 5 2 6 classification_cost… binary 0.401 5 0.0161 Preprocessor1…
由reprex 包(v2.0.1) 於 2021 年 11 月 3 日創建
聲明:本站的技術帖子網頁,遵循CC BY-SA 4.0協議,如果您需要轉載,請注明本站網址或者原文地址。任何問題請咨詢:yoyou2525@163.com.