5
$\begingroup$

I have a situation where we have a number of quantitative features / variables (p) than the number of samples (n). My objective is to classify these samples into groups (may be hierarchical). I can see a good discussion about this in this Q/A post, here at CV. I am aware of discussion about clustering based on high-dimensional data in wiki and its needs.

Here is data example for workout:

set.seed(123)
# matrix of X variable 
xmat <- matrix(sample(-1:1, 2000000, replace = TRUE), ncol = 10000)
colnames(xmat) <- paste ("M", 1:10000, sep ="")
rownames(xmat) <- paste("sample", 1:200, sep = "")

Here are my questions:

  1. What would be best approach ?

  2. I am interested to find implementation codes for a suitable method (may be Subspace clustering or Projected clustering or Correlation clustering or Hybrid approaches) for my case.

$\endgroup$
12
$\begingroup$

First some background:

R is good choice and have so many clustering methods in different packages. The functions include Hierarchical Clustering, Partitioning Clustering, Model-Based Clustering, and Cluster-wise Regression.

Connectivity based clustering or Hierarchical clustering (also called hierarchical cluster analysis or HCA) is a method of cluster analysis which seeks to build a hierarchy of clusters. Strategies for hierarchical clustering generally fall into two types:

  • Agglomerative: This is a "bottom up" approach: each observation starts in its own cluster, and pairs of clusters are merged as one moves up the hierarchy.
  • Divisive: This is a "top down" approach: all observations start in one cluster, and splits are performed recursively as one moves down the hierarchy. In this method, at different distances, different clusters will form, which can be represented using a dendrogram.

Centroid-based clustering: In centroid-based clustering, clusters are represented by a central vector, which may not necessarily be a member of the data set. When the number of clusters is fixed to k, k-means clustering gives a formal definition as an optimization problem: find the k cluster centers and assign the objects to the nearest cluster center, such that the squared distances from the cluster are minimized.

km <- kmeans(iris[,1:4], 3)
plot(iris[,1], iris[,2], col=km$cluster)
    points(km$centers[,c(1,2)], col=1:3, pch=8, cex=2)
table(km$cluster, iris$Species)

Distribution-based clustering: the clustering model most closely related to statistics is based on distribution models. Clusters can then easily be defined as objects belonging most likely to the same distribution. The problem is overfitting.

Density-based clustering: n density-based clustering,[8] clusters are defined as areas of higher density than the remainder of the data set. Objects in these sparse areas - that are required to separate clusters - are usually considered to be noise and border points.

In density based cluster, a cluster is extend along the density distribution. Two parameters is important: "eps" defines the radius of neighborhood of each point, and "minpts" is the number of neighbors within my "eps" radius. The basic algorithm called DBscan proceeds as follows:

First scan: For each point, compute the distance with all other points. Increment a neighbor count if it is smaller than "eps".

Second scan: For each point, mark it as a core point if its neighbor count is greater than "mints"

Third scan: For each core point, if it is not already assigned a cluster, create a new cluster and assign that to this core point as well as all of its neighbors within "eps" radius.

Unlike other cluster, density based cluster can have some outliers (data points that doesn't belong to any clusters). On the other hand, it can detect cluster of arbitrary shapes (doesn't have to be circular at all).

library(fpc)
# eps is radius of neighborhood, MinPts is no of neighbors
# within eps
cluster <- dbscan(sampleiris[,-5], eps=0.6, MinPts=4)
plot(cluster, sampleiris)
plot(cluster, sampleiris[,c(1,4)])
# Notice points in cluster 0 are unassigned outliers
table(cluster$cluster, sampleiris$Species)

With the recent need to process larger and larger data sets (also known as big data), the willingness to trade semantic meaning of the generated clusters for performance has been increasing. This led to the development of pre-clustering methods such as canopy clustering, which can process huge data sets efficiently, but the resulting "clusters" are merely a rough pre-partitioning of the data set to then analyze the partitions with existing slower methods such as k-means clustering.

For high-dimensional data, many of the existing methods fail due to the curse of dimensionality, which renders particular distance functions problematic in high-dimensional spaces. This led to new clustering algorithms for high-dimensional data that focus on subspace clustering (where only some attributes are used, and cluster models include the relevant attributes for the cluster) and correlation clustering that also looks for arbitrary rotated ("correlated") subspace clusters that can be modeled by giving a correlation of their attributes. The three clustering algorithms include PROCLUS, P3C and STATPC.

To your question:

The package Package ‘orclus’ is available to perform subspace clustering and classification. The following is example from the manual:

    # definition of a function for parameterized data simulation
        sim.orclus <- function(k = 3, nk = 100, d = 10, l = 4, sd.cl = 0.05, sd.rest = 1, locshift = 1){
              ### input parameters for data generation
        # k # nk # d # l
        # sd.cl
        # sd.rest # locshift
          number of clusters
          observations per cluster
          original dimension of the data
          subspace dimension where the clusters are concentrated
        (univariate) standard deviations for data generation (within cluster subspace concentration) standard deviations in the remaining space
        parameter of a uniform distribution to sample different cluster means
        x <- NULL
        for(i in 1:k){
        # cluster centers
        apts <- locshift*matrix(runif(l*k), ncol = l)
        # sample points in original space
        xi.original <- cbind(matrix(rnorm(nk * l, sd = sd.cl), ncol=l) + matrix(rep(apts[i,], nk), ncol = l, byrow = TRUE)
     matrix(rnorm(nk * (d-l), sd = sd.rest), ncol = (d-l)))
    # subspace generation
    sym.mat <- matrix(nrow=d, ncol=d)
    for(m in 1:d){
      for(n in 1:m){
        sym.mat[m,n] <- sym.mat[n,m] <- runif(1)
        }
      }
    subspace <- eigen(sym.mat)$vectors
        # transformation
        xi.transformed <- xi.original %*% subspace
        x <- rbind(x, xi.transformed)
        }
        clids <- rep(1:k, each = nk)
        result <- list(x = x, cluster = clids)
        return(result)
        }
    # simulate data of 2 classes where class 1 consists of 2 subclasses
    simdata <- sim.orclus(k = 3, nk = 200, d = 15, l = 4, sd.cl = 0.05, sd.rest = 1, locshift = 1)
      x <- simdata$x
  y <- c(rep(1,400), rep(2,200))
  res <- orclass(x, y, k = 3, l = 4, k0 = 15, a = 0.75)
  res
  # compare results
  table(res$predict.train$class, y)

You may also be interested in HDclassif (An R Package for Model-Based Clustering and Discriminant Analysis of High-Dimensional Data).

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.