pam: K-Medoids Clustering.

In Fuzzy-Logix/AdapteR: Converts R syntax to SQL to provide fast analytics using in-database C++ library DB Lytix.

Description

pam performs k-medoids clustering on FLTable objects.

Usage

pam(x, k, ...)

Arguments

x	an object of class FLTable, can be wide or deep table
k	the number of clusters
diss	logical if x is dissimilarity matrix
metric	only "euclidean" distance supported currently
medoids	initial medoids. Not used currently.
Stand	logical indicating if standardization should be done before calculating diss matrix.
cluster.only	logical if only clustering vector is needed as output
do.swap	logical indicating if swap phase is needed. currently always TRUE
keep.diss	logicals indicating if the dissimilarities and/or input data x should be kept in the result
keep.data	logicals indicating if the dissimilarities and/or input data x should be kept in the result
pomance	logical or integer in 0:2 specifying algorithmic short cuts. currently always FALSE
trace.lev	integer specifying a trace level for printing diagnostics during the build and swap phase of the algorithm. currently always 0
iter.max	the maximum number of iterations allowed
excludeCols	the comma separated character string of columns to be excluded
classSpec	list describing the categorical dummy variables
whereconditions	takes the where_clause as a string
distTable	name of the in-database table having dissimilarity matrix or distance table

Details

The DB Lytix function called is FLKMedoids. K-Medoids clusters the training data. The algorithm used is PAM (Partitioning Around Medoids).

Value

pam gives a list which replicates equivalent R output from pam in cluster package. The mapping table can be viewed using object$mapping if input is wide table.

Constraints

Plotting time increases with size of data and even fail for large datasets because data is fetched to R session for plotting. If classSpec is not specified, the categorical variables are excluded from analysis by default.

See Also

pam for R function reference implementation.

Examples

widetable  <- FLTable("iris", "rownames")
kmedoidsobject <- pam(widetable,3)
print(kmedoidsobject)
plot(kmedoidsobject)
One can specify ClassSpec and transform categorical variables 
before clustering. This increases the number of variables in the plot
because categorical variable is split into binary numerical variables.
The clusters may not be well-defined as is observed in the case below:-
widetable  <- FLTable("iris", "obsid")
pamobjectnew <- pam(widetable,3,classSpec=list("Species(setosa)"))
plot(pamobjectnew)

Fuzzy-Logix/AdapteR documentation built on May 6, 2019, 5:07 p.m.