body-fat-projection-methods.R

library(tidyverse)
library(janitor)
library(tidymodels)
library(plotly)

theme_set(theme_bw())

body <- 
  read_csv("http://staff.pubhealth.ku.dk/~tag/Teaching/share/data/Bodyfat.csv") %>% 
  janitor::clean_names() %>% 
  mutate(subject = format(row_number()))

basic_rec <- 
  recipe(density + bodyfat ~ ., data = body) %>% 
  update_role(subject, new_role = "id") %>% 
  step_normalize(all_predictors())

pca_rec <- 
  basic_rec %>% 
  step_pca(all_predictors(), threshold = 0.95)


make_interactive <- function(p) {
  girafe(ggobj = p)
}

pca_data <- 
  pca_rec %>% 
  prep() %>% 
  juice()

pca_range <- extendrange(c(pca_data$PC1, pca_data$PC2))

pca_plot <- 
  ggplot(pca_data, aes(x = PC1, y = PC2)) + 
  geom_point(aes(text = subject), alpha = .3) + 
  lims(x = pca_range, y = pca_range)
ggplotly(pca_plot)

ica_rec <- 
  basic_rec %>% 
  step_ica(all_predictors(), num_comp = 2)

ica_data <- 
  ica_rec %>% 
  prep() %>% 
  juice() 

ica_range <- extendrange(c(ica_data$IC1, ica_data$IC2))

ica_plot <- 
  ggplot(ica_data, aes(x = IC1, y = IC2)) + 
  geom_point(aes(text = subject), alpha = .3) + 
  lims(x = ica_range, y = ica_range)
ggplotly(ica_plot)


pls_rec <- 
  basic_rec %>% 
  step_pls(all_predictors(), outcome = "density", num_comp = 2)


pls_data <- 
  pls_rec %>% 
  prep() %>% 
  juice() 

pls_range <- extendrange(c(pls_data$IC1, pls_data$IC2))

pls_plot <- 
  ggplot(pls_data, aes(x = PLS1, y = PLS2)) + 
  geom_point(aes(text = subject), alpha = .3) + 
  lims(x = ica_range, y = ica_range)
ggplotly(pls_plot)