UMAP
Also: Uniform Manifold Approximation and Projection
A technique that squashes high-dimensional vectors down to 2 or 3 dimensions for visualisation.
Last updated
Uniform Manifold Approximation and Projection. UMAP takes high- dimensional vectors (embeddings) and squashes them down to 2 or 3 dimensions for visualisation, trying to preserve neighbourhood structure from the original space.
Why it matters for your research. UMAP’s 2-D scatterplot is what people usually mean by “an embedding space”. It is genuinely useful for orientation, and genuinely misleading if read as distance — two points that look far apart on the plot may be close in the true 384- dimensional space, and vice versa. Use UMAP for a vibe check, not for measurements.
In Archēglyph. Lives behind an “Advanced” toggle on the cluster view. Text-first results come from HDBSCAN on the full embeddings; UMAP is offered as visual context only.
Not to be confused with. PCA and t-SNE are alternative projection methods with different trade-offs. All three distort the original geometry; none of them are the clusters.