k-means++

Also written: kmeans++

The default way scikit-learn picks k-means' starting centers. Each new center is chosen with a probability that grows with its distance from those already picked, so starts are spread out and bad runs are less likely.