EasyDeepLearn

Top t-SNE pitfalls to avoid.

hard

Answer

  • (1) Cluster sizes in the map are not meaningful — t-SNE inflates dense clusters.
  • (2) Inter-cluster distances are NOT preserved — 'far' clusters aren't necessarily far in data space.
  • (3) Random-seed sensitivity → run multiple times.
  • (4) Only meaningful in 2/3D.
  • (5) NEVER feed t-SNE embedding into a downstream model — use PCA or UMAP instead.
  • Wattenberg's 'How to Use t-SNE Effectively' is required reading.
Check yourself — multiple choice
  • Distances are meaningful
  • Cluster sizes / inter-cluster distances NOT meaningful; seed-sensitive; only 2/3D; NEVER feed into downstream model — use UMAP/PCA
  • Random
  • Only for regression

t-SNE pitfalls: sizes/distances not meaningful; never as ML input.

#dimensionality-reduction#visualization

Practise Unsupervised Learning

214 interview questions in this topic.

Related questions