What conclusions can you not draw from a t-SNE plot?
hardAnswer
- Cluster sizes are meaningless, because t-SNE expands sparse regions and compresses dense ones to fit everything into two dimensions, so a visually large blob is not a numerous group.
- Distances between clusters are also unreliable: the method preserves local neighbourhoods and deliberately sacrifices global geometry, so two well-separated blobs may be closer in the original space than they look.
- Apparent gaps can be artefacts of the perplexity setting, and running it twice with different seeds gives different layouts.
- What you can read is which points are neighbours of which.
- Treat it as a qualitative sanity check, never as evidence for the number of clusters, and use UMAP if you need somewhat better global structure.
Check yourself — multiple choice
- It gives reliable distances and sizes
- Cluster sizes and between-cluster distances are not meaningful, gaps can be perplexity artefacts, and layouts vary by seed — only local neighbourhoods are trustworthy
- It proves the number of clusters
- It is a linear projection
t-SNE preserves local neighbourhoods at the deliberate expense of global geometry and scale.
#dimensionality-reduction#visualization
Practise Unsupervised Learning
214 interview questions in this topic.