We can now test the Scikit-Learn Isomap implementation using the Olivetti faces dataset (provided by AT&T Laboratories, Cambridge), which is made up of 400 64 × 64 grayscale portraits belonging to 40 different people. Examples of these images are shown here:
The original dimensionality is 4096, but we want to visualize the dataset in two dimensions. It's important to understand that using the Euclidean distance for measuring the similarity of images might not the best choice, and it's surprising to see how well the samples are clustered by such a simple algorithm.
The first step is loading ...