You did well in the first section! Now it's time to learn a new algorithm!

So, unlike centroid, medoid has to be a data point - which is the key difference. In the case of large and 'grouped' data there most likely will be an insignificant difference.

The K-Medoids algorithm works similarly to the K-Means, but in the first step, the random points have to be data points. These are medoids. Then, each data point is associated with the closest medoid by some metric. Then, the medoids swap with some data points, and the cost function (sum of the variances for all the points within a cluster) compares with the one in the previous step until we get the minimum possible values.

How to implement the K-Medoids algorithm using Python? The same way you did with the K-Means algorithm! The key function there is KMedoids from the sklearn_extra.cluster library. The algorithm is the next:

1. Create a KMedoids model assigned to a certain variable.
2. Compute K-Medoids clustering using the .fit() method of KMedoids object with data set as a parameter.
3. Predict the labels using the fitted model by applying the .predict() function to the KMedoids object with the data set as a parameter.
4. (not necessary) Visualize the result of clustering.

For example, let's try to cluster the points using the K-Medoids method. The scatter plot for the data points is below.