K-means Clustering Lab

Find closest centroid

• Run this code for as many times as there are data points in X (notice how grabbing row variable from X.shape)
  • With said data point in mind, one cluster at a time:
    • calculate the delta between the point X and its centroid vectors, for however many dimensions as there are; sum them all
    • appendix that delta to distance array
  • Look at all the distances it gathered over each iteration, and returns the index of the smallest value (which relates to the centroid it was calculated via)
  • Clear the distance array

Compute new centroid location

• For every cluster
  • Gets all data points from the matrix X for said centroid (starting with 0)
  • for said centroid, takes all the data that has been moved into the points matrix and takes the average, one axis at a time (hence axis =0)

Run the k-means algorithm

Can also randomly initialize the centroids:

Image compression with K means

• First two dimensions are location of pixel, then this pixel has separate R,G, and B values
  • For example, original_img[50, 33, 2] gives the blue intensity of the pixel at row 50 and column 33.
• Take all the pictures' RGB values - in 3-dimensional space - and then find optimal centroid clusters. Then replace those pixels with their associated centroid.

Anomaly Detection

Define the values and shape of each feature