Question

When using Kmeans in Weka, one can call getAssignments() on the resulting output of the model to get the cluster assignment for each given instance. Here’s a (truncated) Jython example:

>>>import weka.clusterers.SimpleKMeans as kmeans
>>>kmeans.buildClusterer(data)
>>>assignments = kmeans.getAssignments()
>>>assignments
>>>array('i',[14, 16, 0, 0, 0, 0, 16,...])

The index of each cluster number corresponds to the instance. So, instance 0 is in cluster 14, instance 1 is in cluster 16, and so on.

My question is: Is there something similar for Xmeans? I’ve gone through the entire API here and don’t see anything like that.

Answer 1

Here’s a reply to my question from the Weka listserv:

 "Not as such. But all clusterers have a clusterInstance() method. You can 
 pass each training instance through the trained clustering model to 
 obtain the cluster index for each."

Here’s my Jython implementation of this suggestion:

 >>> import java.io.FileReader as FileReader
 >>> import weka.core.Instances as Instances
 >>> import weka.clusterers.XMeans as xmeans
 >>> import java.io.BufferedReader as read
 >>> import java.io.FileReader
 >>> import java.io.File
 >>> read = read(FileReader("some arff file"))
 >>> data = Instances(read)
 >>> file = FileReader("some arff file")
 >>> data = Instances(file)
 >>> xmeans = xmeans()
 >>> xmeans.setMaxNumClusters(100)  
 >>> xmeans.setMinNumClusters(2) 
 >>> xmeans.buildClusterer(data)# here's our model 
 >>> enumerated_instances = data.enumerateInstances() #get the index of each instance 
 >>> for index, instance in enumerate(enumerated_instances):
         cluster_num = xmeans.clusterInstance(instance) #pass each instance through the model
         print "instance # ",index,"is in cluster ", cluster_num #pretty print results

 instance # 0 is in cluster  1
 instance # 1 is in cluster  1
 instance # 2 is in cluster  0
 instance # 3 is in cluster  0

I’m leaving all of this up as a reference, since the same approach could be use to get cluster assignments for the results of any of Weka’s clusterers.