separate functions for transfer stages

R/kmeans/mapper.R

@@ -0,0 +1,64 @@
+#
+# K-means clustering algorithm
+# Based on Hartigan, Wong (1979)
+#
+
+# Optimal-transfer and quick-transfer stage functions
+source("optra.R")
+source("qtran.R")
+
+# Global variables
+A <- matrix(rnorm(50*2), 50, 2) # Raw data
+M <- length(A[,1]) # Number of points
+N <- length(A[1,]) # Dimensions
+K <- 5 # Number of clusters 
+C <- matrix(0, K, N) # Cluster centres
+IC1 <- rep(0, M) # 1st closest neighbor
+IC2 <- rep(0, M) # 2nd closest neighbor
+NC <- rep(0, K) # Number of points in each cluster
+ITER <- 10 # Iterations
+
+# Euclidean distance
+distance <- function(a, b) {
+  dist <- 0
+  for(i in 1:length(a)) {
+    dist <- dist + (a[i]-b[i])^2
+  }
+  return(sqrt(dist))
+}
+
+main<-function() {
+  # Initialize the cluster centers
+  for(k in 1:K) {
+    for(n in 1:N) {
+      C[k,n] <<- runif(1, min(A[,n]), max(A[,n]))
+    }
+  }
+
+  # Step 1 - Closest neighbors  
+  for(I in 1:M) {
+    d <- rep(0, K)
+    for(k in 1:K) {
+      d[k] <- distance(A[I,], C[k,])
+    }
+    IC1[I] <<- order(d)[1]
+    IC2[I] <<- order(d)[2]   
+  }
+
+  # Step 2 - Update cluster centres
+  for(k in unique(IC1)) {
+    for(n in 1:N) {
+      C[k,n] <<- mean(A[,n][IC1==k])
+    }
+  }
+
+  # Step 3 
+  # Clusters already initiated into the live set
+
+  # Step 4
+  optra(rep(TRUE, K))
+  print(IC1)
+}
+
+# Start the app
+main()