The MNIST Dataset

Now, we will explore SNE and t-SNE using the MNIST dataset provided with the accompanying source code as the basis of our practical examples. Before we continue, we will quickly review MNIST and the data that is within it. The complete MNIST dataset is a collection of 60,000 training and 10,000 test examples of handwritten digits of the numbers 0 to 9, represented as black and white (or grayscale) images that are 28 x 28 pixels in size (giving 784 dimensions or features) with equal numbers of each type of digit (or class) in the dataset. Due to its size and the quality of the data, MNIST has become one of the quintessential datasets in machine learning, often being used as the reference dataset for many research papers in machine learning. One of the advantages of using MNIST to explore SNE and t-SNE compared to other datasets is that while the samples contain a high number of dimensions, they can be visualized even after dimensionality reduction because they can...