Hands-On Machine Learning with TensorFlow.js

Evaluation

There is nothing difficult about evaluating the results of the K-means algorithm. Simply using the loss function provides us with insight so that we can evaluate the clustering result. The loss function is the sum of the squared error:

One challenge that the K-means algorithm usually faces is the decision of K. The desired number of clusters, K, must be provided as a hyperparameter. However, it is difficult to decide on the correct K parameter beforehand. Often, we don't have any information or knowledge on how many data points can be separated into clusters. In this case, the SSE metric is useful. Trying several patterns in a brute ...