Fig. 4From: DiviK: divisive intelligent K-means for hands-free unsupervised clustering in big biological dataFlowchart of K-means initialization procedure. We find a moderately outlying KD-Tree leaf centroid as the first seed and follow an iterative deterministic algorithm similar to the K-means++ algorithm. KD-Tree keeps the order of computational complexity low, while the linear model makes it more predictable than the K-means++ algorithmBack to article page