Online K-means, SGD variant and Mini-batch K-means

1. Online K-means

step1: Initialize the  cluster centers ,......, in any way;

          Create counters ,......, and initialize them to zero;

Step2: Loop

           Get new data point ;

           Determine the closest center to ;

           Update the number of points in that cluster: ;

           Update the cluster center: ;

           End loop

2.  SGD variant

 

3. Mini-batch K-means

    Initialization: mini-batch size , initialize  cluster centers ,......,  picked randomly from dataset

    Loop for   iterations

    For each batch ， examples pickes randomly from :

    For :

         Get the closest center:    to ;

         Update the center count:  ;

         Update the cluster center:

    End for

    End loop

 

 

Reference

1. Onlinek-Means Clustering

2. Web-Scale K-Means Clustering

3. Convergence properties of the k means algorithm