clusterMatch: clusterMatch

In fastLink: Fast Probabilistic Record Linkage with Missing Data

clusterMatch

Description

Creates properly sized clusters for matching, using either alphabetical or word embedding clustering. If using word embedding, the function first creates a word embedding out of the provided vectors, and then runs PCA on the matrix. It then takes the first k dimensions (where k is provided by the user) and k-means is run on that matrix to get the clusters.

Usage

clusterMatch(vecA, vecB, nclusters, max.n, word.embed, min.var, iter.max)

Arguments

vecA	The character vector from dataset A
vecB	The character vector from dataset B
nclusters	The number of clusters to create from the provided data. Either nclusters = NULL or max.n = NULL.
max.n	The maximum size of either dataset A or dataset B in the largest cluster. Either nclusters = NULL or max.n = NULL
word.embed	Whether to use word embedding clustering. Default is FALSE.
min.var	The minimum amount of explained variance (maximum = 1) a PCA dimension can provide in order to be included in k-means clustering when using word embedding. Default is .20.
iter.max	Maximum number of iterations for the k-means algorithm.

Value

clusterMatch returns a list of length 3:

clusterA	The cluster assignments for dataset A
clusterB	The cluster assignments for dataset B
n.clusters	The number of clusters created
kmeans	The k-means object output.
pca	The PCA object output.
dims.pca	The number of dimensions from PCA used for the k-means clustering.

Author(s)

Ben Fifield <[e-mail]>

Examples

data(samplematch)
cl <- clusterMatch(dfA$firstname, dfB$firstname, nclusters = 3)

fastLink documentation built on Nov. 17, 2023, 9:06 a.m.