In-class worksheet 10

Feb 15, 2018

In this worksheet, we will use the library tidyverse:

library(tidyverse)
theme_set(theme_bw(base_size=12)) # set default ggplot2 theme

1. k-means clustering of the iris data set

We can do k-means clustering in R using the function kmeans(). The primary argument to this function is centers, which tells the function how many cluster centers we want to use:

iris %>% select(-Species) %>% # remove Species column
  kmeans(centers=3) ->        # do k-means clustering with 3 centers
  km                          # store result as `km`

# now display the results from the analysis
km
## K-means clustering with 3 clusters of sizes 21, 33, 96
## 
## Cluster means:
##   Sepal.Length Sepal.Width Petal.Length Petal.Width
## 1     4.738095    2.904762     1.790476   0.3523810
## 2     5.175758    3.624242     1.472727   0.2727273
## 3     6.314583    2.895833     4.973958   1.7031250
## 
## Clustering vector:
##   [1] 2 1 1 1 2 2 2 2 1 1 2 2 1 1 2 2 2 2 2 2 2 2 2 2 1 1 2 2 2 1 1 2 2 2 1
##  [36] 2 2 2 1 2 2 1 1 2 2 1 2 1 2 2 3 3 3 3 3 3 3 1 3 3 1 3 3 3 3 3 3 3 3 3
##  [71] 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 1 3 3 3 3 1 3 3 3 3 3 3
## [106] 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
## [141] 3 3 3 3 3 3 3 3 3 3
## 
## Within cluster sum of squares by cluster:
## [1]  17.669524   6.432121 118.651875
##  (between_SS / total_SS =  79.0 %)
## 
## Available components:
## 
## [1] "cluster"      "centers"      "totss"        "withinss"    
## [5] "tot.withinss" "betweenss"    "size"         "iter"        
## [9] "ifault"

The cluster means give us the locations of the centroids of the three clusters. We can access them directly via km$centers:

km$centers
##   Sepal.Length Sepal.Width Petal.Length Petal.Width
## 1     4.738095    2.904762     1.790476   0.3523810
## 2     5.175758    3.624242     1.472727   0.2727273
## 3     6.314583    2.895833     4.973958   1.7031250

The clustering vector tells us for each data point to which cluster it belongs. We can access this vector directly via km$cluster:

km$cluster
##   [1] 2 1 1 1 2 2 2 2 1 1 2 2 1 1 2 2 2 2 2 2 2 2 2 2 1 1 2 2 2 1 1 2 2 2 1
##  [36] 2 2 2 1 2 2 1 1 2 2 1 2 1 2 2 3 3 3 3 3 3 3 1 3 3 1 3 3 3 3 3 3 3 3 3
##  [71] 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 1 3 3 3 3 1 3 3 3 3 3 3
## [106] 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
## [141] 3 3 3 3 3 3 3 3 3 3

We can add this vector to the iris data frame and then plot, coloring by cluster and using different symbol shapes according to species, for comparison:

# we use `factor(km$cluster)` to tell R that the cluster numbers represent distinct categories,
# not continuous values 
iris_clustered <- data.frame(iris, cluster=factor(km$cluster))
ggplot(iris_clustered, aes(x=Petal.Width, y=Sepal.Width, color=cluster, shape=Species)) + geom_point()

Let’s do everything once more:

iris %>% select(-Species) %>% kmeans(centers=3) -> km
iris_clustered <- data.frame(iris, cluster=factor(km$cluster))
ggplot(iris_clustered, aes(x=Petal.Width, y=Sepal.Width, color=cluster, shape=Species)) + geom_point()

You see that the results do not look the same. This happens because the k-means algorithm uses a random starting point. To make the results less random, we can set the parameter nstart to a value other than 1, e.g. nstart=10:

iris %>% select(-Species) %>% kmeans(centers=3, nstart=10) -> km
iris_clustered <- data.frame(iris, cluster=factor(km$cluster))
ggplot(iris_clustered, aes(x=Petal.Width, y=Sepal.Width, color=cluster, shape=Species)) + geom_point()

To determine the best number of clusters in a systematic way, we can plot the within-groups sum of squares against the number of clusters and look for a characteristic slowdown in the decline:

iris_numeric <- select(iris, -Species) # make a data table with only the numeric measurements from iris
# calculate the weighted sum squares (wss) for a single cluster
wss <- (nrow(iris_numeric)-1)*sum(apply(iris_numeric, 2, var))
# now calculate the weighted sum squares for all cluster numbers from 2 to 15
for (i in 2:15) wss[i] <- sum(kmeans(iris_numeric,
                                       nstart=10,
                                       centers=i)$withinss)
# turn wss data into a data frame for plotting
wss_data <- data.frame(centers=1:15, wss)
# make plot
ggplot(wss_data, aes(x=centers, y=wss)) + geom_point() + geom_line() +
  xlab("Number of Clusters") + ylab("Within groups sum of squares")

We see that around 3 clusters, the decline slows down a lot. We would then choose 3 or 4 clusters for this data set.

For more information about choosing the number of clusters, read this post on stackoverflow.com.

2. Clustering the biopsy data set

Take the biopsy data set, cluster it into 2, 3, or 4 clusters, and display the clusters in principal-component space. (In other words, plot PC2 vs PC1 and color by cluster. You can set the plotting symbol by outcome, to see how clustering compares with known outcomes.)

# load in the biopsy data set
biopsy <- read.csv("http://wilkelab.org/classes/SDS348/data_sets/biopsy.csv")
head(biopsy)
##   clump_thickness uniform_cell_size uniform_cell_shape marg_adhesion
## 1               5                 1                  1             1
## 2               5                 4                  4             5
## 3               3                 1                  1             1
## 4               6                 8                  8             1
## 5               4                 1                  1             3
## 6               8                10                 10             8
##   epithelial_cell_size bare_nuclei bland_chromatin normal_nucleoli mitoses
## 1                    2           1               3               1       1
## 2                    7          10               3               2       1
## 3                    2           2               3               1       1
## 4                    3           4               3               7       1
## 5                    2           1               3               1       1
## 6                    7          10               9               7       1
##     outcome
## 1    benign
## 2    benign
## 3    benign
## 4    benign
## 5    benign
## 6 malignant
# do PCA on biopsy data set
biopsy %>% select(-outcome) %>% scale() %>% prcomp() -> pca

Hint: To create the plot, you will have to make a new data frame out of the pca coordinates (pca$x) and the clustering results (km$cluster).

# your R code goes here

3. If this was easy

For the biopsy dataset, plot the within-group sum of squares against the number of clusters, to determine the appropriate number of clusters.

# your R code goes here

Now, go back to the original clustering of the iris data set, with 3 clusters:

iris %>% select(-Species) %>% kmeans(centers=3, nstart=10) -> km
iris_clustered <- data.frame(iris, cluster=factor(km$cluster))
ggplot(iris_clustered, aes(x=Petal.Width, y=Sepal.Width, color=cluster, shape=Species)) + geom_point()