The KMeans implementation has the following properties. The shape of the input data is assumed to be [batch size, dimension]. This would be [90, 2] in this case:
export class KMeans { // The desired number of clusters k: number; // The dimension of data points. dim: number; centroids: tf.Tensor; // Given data points xs: tf.Tensor; clusterAssignment: tf.Tensor; constructor(xs: tf.Tensor, k: number) { this.dim = xs.shape[1]; this.k = k; // Initialize centroids by picking up K random points. this.centroids = tf.randomNormal([this.k, this.dim]); this.xs = xs; }}
clusterAssignment is a tensor that stores the index of the cluster that each data point is assigned to. Thus, its shape will be [N, 1]. N is the number of data ...
