A very simple implementation of k-means clustering.


A clustering session typically involves:

  • Setting a distance measureme to use.
distanceMeasure = DistanceType.squaredEuclidian; // default
  • Creating a List of Instances. This is generally done by mapping a list of whatever data structures is available.
// For example, data might be a List<String> such
// that each String represents an individual instance.
List<Instance> instances = data.map((datum) {
  List<num> coordinates = ...;
  String id = ...;
  return new Instance(coordinates, id: id); 
  • Creating a List of Clusters. This can be done manually (e.g. create a set of randomly placed clusters). A convenience function initialClusters exists that takes in the list of Instances already created and randomly generates clusters from the instances such that instances more distant to the previous cluster are more likely to seed the next cluster.
List<Cluster> clusters = initialClusters(3, instances, seed: 0);
  • Running the algorithm using the kmeans function. This is a side-effect heavy function that iteratively shifts the clusters towards the mean position of the associated instances and reassigns instances to the nearest cluster.
kmeans(clusters: clusters, instances: instances);
  • Inspecting the instances property of each cluster.
clusters.forEach((cluster) {
  cluster.instances.forEach((instance) {
    print("  - $instance");

Please see the associated wiki for more details and examples.

Please file feature requests and bugs at the issue tracker.


A very simple implementation of k-means clustering.