tSNE
A nonlinear dimensionality reduction and visualization algorithm that minimizes the KL divergence between high-dimensional and low-dimensional probability distributions, drawing similar samples together while pushing dissimilar ones apart. It is effective at revealing cluster-like local structures and is widely used for high-dimensional data visualization.
Timeline (last 90 days)
The author tried three dimensionality reduction methods—PCA, UMAP, and tSNE—and ultimately concluded that tSNE produced the best structure.
The author simultaneously mapped the dimensionality reduction results to XYZ spatial coordinates and RGB color channels, making each font a colored point where proximity in both position and color represents similarity in visual features.
tSNE is a nonlinear method that excels at preserving local neighborhood structures and revealing cluster structures, but the global distances between different clusters are not strictly comparable.
UMAP differs from tSNE in its mathematical foundations, is faster, and preserves global structure better than tSNE.
The map generated by the author using the Google Fonts corpus has an overall flower-like structure, with the model clustering most cursive/handwritten fonts toward the center of the flower.
All Facts (5)
The author tried three dimensionality reduction methods—PCA, UMAP, and tSNE—and ultimately concluded that tSNE produced the best structure.
50%UnverifiedThe author simultaneously mapped the dimensionality reduction results to XYZ spatial coordinates and RGB color channels, making each font a colored point where proximity in both position and color represents similarity in visual features.
50%UnverifiedtSNE is a nonlinear method that excels at preserving local neighborhood structures and revealing cluster structures, but the global distances between different clusters are not strictly comparable.
50%UnverifiedUMAP differs from tSNE in its mathematical foundations, is faster, and preserves global structure better than tSNE.
50%UnverifiedThe map generated by the author using the Google Fonts corpus has an overall flower-like structure, with the model clustering most cursive/handwritten fonts toward the center of the flower.
50%