The k-means clustering algorithm finds centroids that best represent the data using an iterative process. The algorithm starts with a predefined set of centroids, which are normally data points taken from the training data. The k in k-means is the number of centroids to look for and how many clusters the algorithm will find. For instance, setting k to 3 will find three clusters in the dataset.
There are two phases to the k-means: assignment and updating. They are explained as below:
- In the assignment step, we set a label to every sample in the dataset linking it to the nearest centroid. For each sample nearest to centroid 1, we assign the label 1. For each sample nearest to centroid 2, we assign a label 2 and so on for each of the k centroids. These labels form the clusters, so we say that each data point with the label 1 is in cluster 1 (at this time only, as assignments can change as the algorithm runs).
- In the updating step, we take each of the clusters and compute...