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