kmeans_example.Rmd

---

title: "cluster"
output: github_document

---

knitr::opts_chunk$set(echo = TRUE)

library(tidyverse)
library(broom)
library(gganimate)
library(gridExtra)

## Code to do each iteration of Kmeans and prepare the data for future plotting
# I Definetly should be able to clean up this code with a loop or function but i cannot think right now...

set.seed(1230)


# Initialize Clustering
iris_sample0 <- iris %>%
  select(x = Sepal.Length, y = Sepal.Width) %>% 
  mutate(iter0 = factor(sample(c(1,2,3), size = n(), replace = TRUE)))

iris_means0 <- iris_sample0 %>% 
  select(group = iter0, x, y) %>%
  group_by(group) %>% 
  summarize(center_x = mean(x), center_y = mean(y)) %>% 
  mutate(iteration = "iter0")


# Perform First Iteration
iris_sample1 <- iris_sample0 %>% 
  mutate(dist1 = sqrt((x - iris_means0$center_x[1])^2 + (y - iris_means0$center_y[1])^2),
         dist2 = sqrt((x - iris_means0$center_x[2])^2 + (y - iris_means0$center_y[2])^2),
         dist3 = sqrt((x - iris_means0$center_x[3])^2 + (y - iris_means0$center_y[3])^2),
         iter1 = case_when(
           pmin(dist1, dist2, dist3) == dist1 ~ 1,
           pmin(dist1, dist2, dist3) == dist2 ~ 2,
           pmin(dist1, dist2, dist3) == dist3 ~ 3),
         iter1 = factor(iter1)) %>% 
  select(-starts_with("dist"))
           
iris_means1 <- iris_sample1 %>% 
  select(group = iter1, x, y) %>%
  group_by(group) %>% 
  summarize(center_x = mean(x), center_y = mean(y)) %>% 
  mutate(iteration = "iter1")


# Perform Second Iteration
iris_sample2 <- iris_sample1 %>% 
  mutate(dist1 = sqrt((x - iris_means1$center_x[1])^2 + (y - iris_means1$center_y[1])^2),
         dist2 = sqrt((x - iris_means1$center_x[2])^2 + (y - iris_means1$center_y[2])^2),
         dist3 = sqrt((x - iris_means1$center_x[3])^2 + (y - iris_means1$center_y[3])^2),
         iter2 = case_when(
           pmin(dist1, dist2, dist3) == dist1 ~ 1,
           pmin(dist1, dist2, dist3) == dist2 ~ 2,
           pmin(dist1, dist2, dist3) == dist3 ~ 3),
         iter2 = factor(iter2)) %>% 
  select(-starts_with("dist"))
           
iris_means2 <- iris_sample2 %>% 
  select(group = iter2, x, y) %>%
  group_by(group) %>% 
  summarize(center_x = mean(x), center_y = mean(y)) %>% 
  mutate(iteration = "iter2")


# Perform Third Iteration
iris_sample3 <- iris_sample2 %>% 
  mutate(dist1 = sqrt((x - iris_means2$center_x[1])^2 + (y - iris_means2$center_y[1])^2),
         dist2 = sqrt((x - iris_means2$center_x[2])^2 + (y - iris_means2$center_y[2])^2),
         dist3 = sqrt((x - iris_means2$center_x[3])^2 + (y - iris_means2$center_y[3])^2),
         iter3 = case_when(
           pmin(dist1, dist2, dist3) == dist1 ~ 1,
           pmin(dist1, dist2, dist3) == dist2 ~ 2,
           pmin(dist1, dist2, dist3) == dist3 ~ 3),
         iter3 = factor(iter3)) %>% 
  select(-starts_with("dist"))
           
iris_means3 <- iris_sample3 %>% 
  select(group = iter3, x, y) %>%
  group_by(group) %>% 
  summarize(center_x = mean(x), center_y = mean(y)) %>% 
  mutate(iteration = "iter3")


# Perform Fourth Iteration
iris_sample4 <- iris_sample3 %>% 
  mutate(dist1 = sqrt((x - iris_means3$center_x[1])^2 + (y - iris_means3$center_y[1])^2),
         dist2 = sqrt((x - iris_means3$center_x[2])^2 + (y - iris_means3$center_y[2])^2),
         dist3 = sqrt((x - iris_means3$center_x[3])^2 + (y - iris_means3$center_y[3])^2),
         iter4 = case_when(
           pmin(dist1, dist2, dist3) == dist1 ~ 1,
           pmin(dist1, dist2, dist3) == dist2 ~ 2,
           pmin(dist1, dist2, dist3) == dist3 ~ 3),
         iter4 = factor(iter4)) %>% 
  select(-starts_with("dist"))
           
iris_means4 <- iris_sample4 %>% 
  select(group = iter4, x, y) %>% 
  group_by(group) %>% 
  summarize(center_x = mean(x), center_y = mean(y)) %>% 
  mutate(iteration = "iter4")

# Perform Fifth Iteration
iris_sample5 <- iris_sample4 %>% 
  mutate(dist1 = sqrt((x - iris_means4$center_x[1])^2 + (y - iris_means4$center_y[1])^2),
         dist2 = sqrt((x - iris_means4$center_x[2])^2 + (y - iris_means4$center_y[2])^2),
         dist3 = sqrt((x - iris_means4$center_x[3])^2 + (y - iris_means4$center_y[3])^2),
         iter5 = case_when(
           pmin(dist1, dist2, dist3) == dist1 ~ 1,
           pmin(dist1, dist2, dist3) == dist2 ~ 2,
           pmin(dist1, dist2, dist3) == dist3 ~ 3),
         iter5 = factor(iter5)) %>% 
  select(-starts_with("dist"))
           
iris_means5 <- iris_sample5 %>% 
  select(group = iter5, x, y) %>% 
  group_by(group) %>% 
  summarize(center_x = mean(x), center_y = mean(y)) %>% 
  mutate(iteration = "iter5")



df_kmeans_example <- iris_sample5
df_kmeans_ex_center <- bind_rows(iris_means0, iris_means1, iris_means2, iris_means3, iris_means4, iris_means5)

# Create facet plot for  K means algorithm


df_kmeans_centers <- df_kmeans_ex_center %>% 
  mutate(sample = 1) %>% 
  bind_rows(df_kmeans_ex_center) %>% 
  replace_na(list(sample = 2)) %>% 
  mutate(iteration = case_when(
            sample == 1 ~ paste(iteration, "b", sep = ""),
            sample == 2 ~ paste("iter", as.numeric(str_sub(iteration, start = -1)) + 1, "a", sep = ""))) %>% 
  select(-sample, cluster = group) %>% 
  filter(iteration != "iter6a")

df_kmeans <- df_kmeans_example %>% 
  rename(iter0a = iter0, iter1a = iter1, iter2a = iter2,
         iter3a = iter3, iter4a = iter4, iter5a = iter5) %>% 
  mutate(iter0b = iter0a, iter1b = iter1a, iter2b = iter2a,
         iter3b = iter3a, iter4b = iter4a, iter5b = iter5a) %>% 
  gather(starts_with("iter"), key = "iteration", value = "group") %>% 
  mutate(iteration = factor(iteration)) %>% 
  rename(cluster = group)


iris_sample0 %>% 
  ggplot(aes(x,y, col = "NA")) +
    geom_point() +
    scale_color_manual(values= c("#000000")) +
    labs(title = "K Means Clustering Algorithm",
         col = "cluster")


df_kmeans %>% 
  filter(str_detect(iteration, "b")) %>% 
  ggplot(aes(x = x, y = y, col = cluster)) +
    geom_point(alpha = 0.6) +
    geom_point(data = filter(df_kmeans_centers, str_detect(iteration, "b")),
               aes(x = center_x, y = center_y, fill = cluster), shape = 21, col = "black",size = 3) +
    facet_wrap(~iteration) +
    labs(title = "K Means Clustering Algorithm")
  

tot <- df_kmeans %>% 
  filter(iteration == "iter5b") %>% 
  summarize(center_x = mean(x), center_y = mean(y)) %>% 
  mutate(cluster = "total")


df_kmeans %>% 
  filter(iteration == "iter5b") %>% 
  ggplot(aes(x = x, y = y, col = cluster)) +
    geom_point(alpha = 0.6) +
    geom_point(data = filter(df_kmeans_centers, iteration == "iter5b"),
               aes(x = center_x, y = center_y, fill = cluster), shape = 21, col = "black",size = 4) +
    geom_point(data = tot,
               aes(x = center_x, y = center_y),
               shape = 3, col = "black",size = 7) +
    labs(title = "K Means Clustering Algorithm")

df <- iris %>%
  select(x = Sepal.Length, y = Sepal.Width)

for(k in 1:9){
 
  model <- kmeans(df, centers = k, nstart = 20) 
  
  clusters <- tibble(k = model$cluster)
  
  df <- df %>% bind_cols(clusters)
}

df %>% 
  gather(-x, -y, key = "Number_of_clusters", value = "Value") %>% 
  mutate(No_of_Clusters = factor(sort(rep(1:9, times = 150)))) %>% 
  ggplot(aes(x, y, col = factor(Value))) +
    geom_point() +
    facet_wrap(~No_of_Clusters) +
    labs(title = "K means models with varying K", 
         col = "Cluster")
  

tibble(k = 1:9) %>%
  mutate(clusts = map(k, ~kmeans(df, .x, nstart = 20)),
         glanced = map(clusts, glance)) %>% 
  unnest(glanced) %>% 
  ggplot(aes(x = k, y = tot.withinss)) +
    geom_line() +
    geom_point() +
    scale_x_continuous(breaks = 1:9) +
    labs(title = "K Means Elbow Plot")

# Create gganimation for kmeans algorithm

df_kmeans %>% 
  mutate(group = seq_len(nrow(df_kmeans))) %>% 
  ggplot(aes(x = x, y = y, col = cluster, group = group)) +
    geom_point(alpha = 0.8) +
    geom_point(data = df_kmeans_centers, aes(x = center_x, y = center_y, fill = cluster, group = cluster), shape = 21, col = "black",size = 6) +
    labs(title = "{closest_state}") +
    transition_states(iteration, transition_length = 5, state_length = 10) +
    enter_fade() + 
    exit_fade() +
    ease_aes('sine-in-out')