Local structure is the pattern of relationships among nearby data points in a representation. Preserving it keeps similar points close after dimensionality reduction. section 7.1.4
PCA can lose local structure when data lies on a nonlinear manifold, as with the Swiss roll. Nonlinear manifold learning can instead recover the underlying layout. section 7.1.4
UMAP preserves local relationships by:
- Calculating distances between points in high-dimensional space. section 7.1.4
- Converting those distances into similarity scores. section 7.1.4
- Moving low-dimensional projections closer together or farther apart according to those scores. section 7.1.4
Chemical-space projection
A UMAP projection of COVID Moonshot compounds can be viewed across measured activity and chemical-series designations, revealing whether nearby compounds share meaningful patterns. section 7.1.4
Figure 7.6 UMAP develops an understanding of local relationships by calculating the distances between each pair of points in high-dimensional space, and using these as the basis for high-dimensional similarity scores that are used to move low-dimensional projects of the points closer or further from one another.

Trace how UMAP converts high-dimensional distances into local relationships in a low-dimensional projection. section 7.1.4
