Parallel Coordinates
by marimo team
Explore Fashion MNIST in higher dimensions with interactive parallel coordinate brushing.
About this notebook
This notebook loads the Fashion MNIST dataset, a set of 28-by-28 grayscale images of clothing, and reduces the 784 pixel values of each image to a small number of components with PCA. A checkbox switches the reduction to UMAP, and sliders set the sample size and the number of components.
The components appear in a parallel coordinates plot from the wigglystuff package, with one line per image, colored by clothing category. Drag a brush along any axis to filter the lines and see how the categories separate.
A final cell reacts to the brushes: it shows the first ten images that pass the current filters, with their category labels, so you can inspect what the selected region contains.
