pix-plot VS troika

Density based clustering with high dimensional data will tend to struggle. This is because, in high enough dimensions, you need a lot of samples to see any density. Also distances start to look very similar (from the curse of dimensionality). To get any traction on such things you need some form of dimension reduction. For something like this non-linear techniques are going to be better. If you want a pipeline of standard parts then something like:

Pretrained-CNN --> UMAP --> HDBSCAN

can turn out relatively reasonable results, especially if the UMAP you use for the clustering is to more than 2 or 3 dimensions (often 5 to 20 is good, depending on the data). You can, of course, still use a 2D UMAP to visualize the results. If you want such a pipeline packaged up then consider the PixPlot package, designed for exactly this use case, from the Yale Digital Humanities Lab: https://github.com/YaleDHLab/pix-plot

* Disclaimer: I am highly biased, as an author of both HDBSCAN and UMAP implementations.

troika (specifically troika-three-text)

Text is relatively hard to render with WebGL and browsers have a limit, of usually 16 framebuffers that you can use for WebGL contexts. Our charts are written for a WebGL target so we don't want to use up that precious resource for text rendering. In our specific use case we usually have a large degree of control over the browser so we can configure it, but if the target is web we don't have that control. Grabbing a WebGL context also takes a significant period of time, usually a hundred milliseconds or more on my computer, and not displaying data in that time isn't a great user experience.

There's a great WebGL library for doing that on the web using any .ttf, .otf, or .woff font - https://github.com/protectwise/troika/tree/master/packages/t...