Logistic Regression • oregonfrogs

Here is an application of the logistic_reg glm with oregonfrogs.

library(oregonfrogs)
library(tidyverse)
library(tidymodels)

frogs <- oregonfrogs %>%
  mutate(survey_date=as.Date(survey_date,"%m/%d/%Y"),
         sex=as.factor(sex))%>%
  arrange(doy)

set.seed(123)
split <- initial_split(frogs, strata = sex, prop = 0.9)
training <- training(split)
test <- testing(split)

cv_folds <- vfold_cv(training,strata = sex,v = 10,repeats = 5)

log_mod <- logistic_reg() %>%
  set_mode("classification") %>%
  set_engine("glm")

library(themis)
recipe <- recipe(sex ~ . ,training) %>% 
    step_downsample(sex) %>%
    step_date(survey_date,keep_original_cols = FALSE) %>% 
    step_corr(all_numeric(),threshold = 0.8) %>%
    step_dummy(all_nominal(), -all_outcomes()) %>%
    step_zv(all_numeric()) %>%
    step_normalize(all_numeric()) 

log_wkf <- workflow() %>%
  add_recipe(recipe) %>%
  add_model(log_mod) %>%
  fit(training)

final_mod <- last_fit(log_wkf, split)

final_mod %>%
  augment() %>%
  conf_mat(truth = sex, estimate = .pred_class)
          Truth
Prediction  0  1
         0  6  7
         1  3 16

two_class_curve <- final_mod %>%
  augment() %>%
  roc_curve(truth = sex, estimate = .pred_0)


autoplot(two_class_curve)