Use t-SNE to reduce the MNIST dataset down to two dimensions and plot the result. You can use a scatter plot using 10 different colors to represent each image's target class. Since dimensionality reduction on the full 60,000 images takes a very long time, you can choose a random subset of 10,000 images. Which digits are easily distinguishable from the others, and which groups of digits are often hard to distinguish from each other? The MNIST dataset can be downloaded using the following code: X, y = sklearn.datasets.fetch_openml ('mnist_784', return_X_y=True). Note that X is a DataFrame. Tip: to speed up the t-SNE computation use the argument n_jobs=-1 in the constructor of TSNE. This will utilize all the available CPUs on your machine for the neighbors search.