Is it feasible to use t-SNE to reduce a dataset to one dimension?

Suppose that I have a matrix, $X$, can I reduce it to a column vector, $Y$ with t-SNE? Suppose that $X$ has 100 columns, how much information can I expect to lose by reducing it to just a column vector with t-SNE (if possible)?

References: L.J.P. van der Maaten and G.E. Hinton. Visualizing High-Dimensional Data Using t-SNE. Journal of Machine Learning Research 9(Nov):2579-2605, 2008. • L.J.P. van der Maaten. Learning a Parametric Embedding by Preserving Local Structure. In Proceedings of the Twelfth International Conference on Artificial Intelligence and Statistics (AI-STATS), JMLR W&CP 5:384-391, 2009. • L.J.P. van der Maaten. Barnes-Hut-SNE. In Proceedings of the International Conference on Learning Representations, 2013. All text available at: http://homepage.tudelft.nl/19j49/t-SNE.html


1 Answer 1


Yes, why not. t-SNE can be used to reduce any dimension to one-dimension. The question about how much information will get lost is not so simple. It is even not clear how to measure or define the information of a high dimensional data.

  • $\begingroup$ Thanks. I intend to use this for clustering. Given enough domain knowledge, I should be able to sense-check the clusters. $\endgroup$
    – power
    Commented Nov 17, 2013 at 22:50
  • 2
    $\begingroup$ I have recently tested tSNE, CCA and RPM algorithms to reduce 1500 dimensional data to one dimension. tSNE worked pretty well, but not the best from a particular point of view (see jamesxli.blogspot.ca/2013/11/on-multidimensional-sorting.html). To make a judgement about the performance, I would trust more domain knowledge or visual checking, than ad-hoc statistics. $\endgroup$
    – James LI
    Commented Nov 17, 2013 at 23:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.