In goodekat/TreeTracer: Trace Plots Using ggplot2

knitr::opts_chunk$set(
  echo = TRUE,
  dpi = 300
)

TreeTracer `r emo::ji("tree")` `r emo::ji("pen")`

The beginnings...

TreeTracer is an R package for creating trace plots of trees from random forests fit using the randomForest R package. Trace plots are useful tools for visually comparing trees from a random forest. See @urbanek:2008 for additional information about trace plots.

# Load packages
library(dplyr)
library(ggpcp)
library(ggplot2)
library(TreeTracer)

Penguin data

# Load the Palmer penguins data
penguins <- na.omit(palmerpenguins::penguins)

# Select the features for training the model
penguin_features <- 
  penguins %>% 
  select(bill_length_mm, bill_depth_mm, flipper_length_mm, body_mass_g)

# Create a parallel coordinate plot of features that will
# be used to fit a random forest colored by the variable 
# of interest to predict (species)
penguins %>%
  ggplot(aes(color = species)) +
  geom_pcp(aes(
    vars = vars(
      bill_length_mm, 
      bill_depth_mm, 
      flipper_length_mm, 
      body_mass_g
    )
  ), alpha = 0.5) + 
  scale_color_brewer(palette = "Paired")

Random forest model

# Fit a random forest
set.seed(71)
penguin_rf <-
  randomForest::randomForest(
    species ~ bill_length_mm + bill_depth_mm + flipper_length_mm + body_mass_g,
    data = penguins, 
    ntree = 50
  )

# Print feature importance
penguin_rf$importance %>% 
  data.frame() %>% 
  arrange(desc(MeanDecreaseGini))

# Trace plots of trees in the forest
trace_plot(
  rf = penguin_rf,
  train = penguin_features,
  tree_ids = 1:penguin_rf$ntree,
  alpha = 0.4
) + 
  theme(aspect.ratio = 1)