Clustering K-means

K-means algorithm step by step:

1) Choose number of K clusters

2) Select random k points as centroids

3) Assign each point to closest(based on distance) centroid and recompute the centroid until last point

4) repeat step 3 until centroid doesn't  change.

K-means random initialization Trap:

What would happen if we did bad random initialization?

If we select different centroids we may get different clusters.

how to tackle this?

K-means++ algorithm.

Within Cluster Sum of Squares(WCSS):

p--> point c- centriod

choose low WCSS. that leads to n(total points) clusters. 

Then how to choose number of clusters?

Elbow method:

4 can be chosen as number of clusters as graph(WCSS) drops after 4

Python implementation:

see the python implementation file in git

https://github.com/hknarra/ML/blob/master/my%20clustering.ipynb