We have now reached the final section of this book, and in this section, we will be talking about low-dimensional representation. We will see what multidimensional scaling (MDS) is, and demonstrate how to perform it.
With MDS, we start with a distance matrix. This could have been computed in any way, using any distance metric we want. Having gotten the distance matrix, we then construct Euclidean coordinates for each point. Perhaps these coordinates representing our data preserve the distances described in the original matrix. If it is not possible, however, we can only hope that the error between the actual and constructed distances is small.
