README.md

In Zelazny7/alamode: K-Modes using Gower Distance

KModes Clustering Using Gower Distance

Most clustering examples use simple datasets with all numeric features. But what if you need to cluster a dataset with mixed features? Gower distance is a distance metric that tries to generalize similarity across different types of features.

x1 <- data.frame(a="cat", b=1, c=TRUE, stringsAsFactors=FALSE)
x2 <- data.frame(a="dog", b=0.9, c=FALSE, stringsAsFactors=FALSE)
x3 <- data.frame(a="goat", b=0.8, c=FALSE, stringsAsFactors=FALSE)

gower_dist(x1, rbind(x1, x2, x3))

## [1] 0.0000000 0.8333333 1.0000000

KModes

Unlike KMeans, KModes uses the modes of each cluster rather than the average. For categorical data, this means taking the value that appears the most often. For numeric data, this is the median value. If there is a tie for the most frequent value, the first one is taken.

cats <- c("cyl","vs","am","gear","carb")
mtcars[cats] <- lapply(mtcars[cats], factor)

km <- kmodes_gower(mtcars, 3)

## Iter   % Changed Cluster
##   1              37.50
##   2              18.75
##   3               6.25
##   4               6.25
##   5               3.12
##   6               0.00

show(km)

## K-modes clustering with 3 clusters of sizes 16, 11, 5
## 
## Cluster modes:
##    mpg cyl   disp  hp drat     wt   qsec vs am gear carb
## 1 22.8   4 120.65  96 3.92 2.5425 18.605  1  1    4    2
## 2 15.0   8 360.00 215 3.15 3.5700 17.020  0  0    3    4
## 3 17.3   8 275.80 180 3.07 3.7300 18.000  0  0    3    3
## 
## Within cluster sum of distances by cluster
## [1] 4.484579 2.259044 1.325879
## 
##  within_dist / total_dist = 72.2 %)

plot(mtcars, col=km$clusters)

Prediction Methods

Once a dataset is cluster with kmodes_gower the resulting object can be used for predictions on new data. There are two kinds of predictions: Clusters and Distance. cluster returns the index of the nearest cluster for each record of the input dataset. distance returns a matrix with K columns each containing the gower distance between the input records and each mode. The argument normalize determines whether the rows of the distance matrix should sum to one.

cluster <- predict(km, mtcars) # default returns clusters
table(cluster)

## cluster
##  1  2  3 
## 21  8  3

d1 <- predict(km, mtcars, type="distance", normalize=FALSE)
d2 <- predict(km, mtcars, type="distance", normalize=TRUE)

head(d1)

##           [,1]      [,2]      [,3]
## [1,] 0.3189645 0.4345832 0.5039349
## [2,] 0.3188312 0.4225952 0.4919469
## [3,] 0.1028999 0.6566431 0.6138735
## [4,] 0.4645546 0.3917196 0.3430850
## [5,] 0.5294238 0.1210936 0.1377230
## [6,] 0.4957737 0.4041956 0.3547231

head(d2)

##            [,1]      [,2]      [,3]
## [1,] 0.25365323 0.3455978 0.4007490
## [2,] 0.25850342 0.3426337 0.3988629
## [3,] 0.07492257 0.4781092 0.4469682
## [4,] 0.38733570 0.3266074 0.2860569
## [5,] 0.67165278 0.1536252 0.1747220
## [6,] 0.39513564 0.3221472 0.2827172

Zelazny7/alamode documentation built on May 17, 2019, 6:09 a.m.