在R中使用CLARA函数计算距离时出错

Question

I am creating a market segmentation of consumers by clustering in 3 categories. I am using the cluster CRAN package with the CLARA clustering algorithm.

The data has 12901 observations with 34 variables taking on ordinal and NA values.

The ordinal values do not have the same increments between categories. For example, in the HouseholdIncome column, the categories are "0-15k", "15k-25k", "25k-35k", "35k-50k", "50k-75k", "75k-100k", "100k-125k", "125k-150k", "150k-175k", "175k-200k", "200k-250k", "250k+".

Every row has a least 1 observation.

> which(rowSums(is.na(Store2df))==ncol(Store2df))
named integer(0)

Here's the first five observations of first seven variables.

> head(Store2df, n=5)
    Age Gender HouseholdIncome MaritalStatus PresenceofChildren HomeOwnerStatus HomeMarketValue
1  <NA>   Male            <NA>          <NA>               <NA>            <NA>            <NA>
2 45-54 Female            <NA>          <NA>               <NA>            <NA>            <NA>
5 45-54 Female        75k-100k       Married                Yes             Own       150k-200k
6 25-34   Male        75k-100k       Married                 No             Own       300k-350k
7 35-44 Female       125k-150k       Married                Yes             Own       250k-300k

Here's the code for the clara function:

> library(cluster)
> #Clara algorithm
> #Set seed for reproducibility
> set.seed(1)
> #Changing medoids.x and keep.data = TRUE - new way 
> client2.clara <- clara(Store2df, 3, metric = "manhattan", stand = FALSE, samples = 5,
+                        sampsize = (2500), medoids.x = TRUE, keep.data = TRUE, 
+                        rngR = TRUE, pamLike = TRUE)
#Error in clara(Store2df, 3, metric = "manhattan", stand = FALSE, samples = 5,  : 
  #Each of the random samples contains objects between which no distance can be computed.

Please let me know if I can provide more information.

Source code for CLARA:

ndyst = as.integer(if(metric == "manhattan") 2 else 1),

Answer 1

Each of the random samples contains objects between which no distance can be computed.

Take this error message seriously...

metric = "manhattan"

is not defined for categorial variables.

Manhattan and Euclidean distances operate on numeric vectors (which also should be linearly scaled, and eg not angles or logscaled).