3rd study

R/decision_tree.R

@@ -0,0 +1,76 @@
+library(tidymodels)
+library(tidyverse)
+library(palmerpenguins)
+library(rpart)
+library(rpart.plot)
+library(magrittr)
+
+set.seed(1234)
+
+penguins %<>% 
+    drop_na() %>% 
+    initial_split(prop = 0.7,
+                  strata = species)
+
+train_data <- penguins %>% training()
+test_data <- penguins %>% testing()
+
+penguins_recipe <- train_data %>% 
+    recipe(species ~ .) %>% 
+    step_naomit(all_numeric(), all_nominal()) %>% 
+    prep()
+
+train_data <- juice(penguins_recipe)
+test_data <- penguins_recipe %>% bake(test_data)
+
+dt_model <-
+    decision_tree(
+        cost_complexity = 0.01, # cp
+        tree_depth = 10, # max depth
+        min_n = 5
+    ) %>% 
+    set_engine("rpart") %>%     
+    set_mode("classification")
+
+dt_model_fit <- 
+    dt_model %>% 
+    fit(species ~ ., data = train_data)
+
+result <- dt_model_fit$fit
+
+rpart.plot(result)
+printcp(initial_fit)
+plotcp(initial_fit)
+
+# ==============================================
+
+model_f <- penguins %>% 
+    select(-species) %>% 
+    names() %>% paste(collapse = " + ") %>%
+    paste("species~", .) %>% 
+    as.formula()
+
+# fitting
+initial_fit <- rpart(model_f,
+             data = penguins, 
+             method = "class", # regression = "anova"
+             control = rpart.control(
+                 minsplit = 5,
+                 minbucket = 2,
+                 cp = 0,
+                 maxdepth = 20),
+             parms = list(split = "information")
+)
+
+rpart.plot(initial_fit)
+printcp(initial_fit)
+plotcp(initial_fit)
+
+
+pdt1 <- prune(initial_fit, cp = initial_fit$cptable[which.min(initial_fit$cptable[, "xerror"]), "CP"])
+
+# Plot the tree
+rpart.plot(pdt1)
+
+
+