K-means question. if we start from the centriod as random selection. Will the results of each calculation always be the same or similar? or very different?

Course Q&A Machine Learning Specialization Unsupervised Learning, Recommenders, Reinforcement
1

[Uploading: image.png…]

for the K-means algothrism, if we start from the centriod as random selection.
Will the results of each calculation always be the same or similar? or very different?

2

If you intended to upload an image, it didn’t work.

Each set of random centroid selections may give different results.

You could try several separate attempts and keep the one that gives the lowest sum of the squared centroid differences - sort of equivalent to a cost function for K-means.