This project is intended as a flexible implementation of t-SNE [1] and dynamic t-SNE [2].

The t-SNE cost function is defined symbolically and automatically translated into efficient (CPU or GPU) code using theano. Because the derivatives are also computed automatically, testing alternative cost functions becomes very easy.

Dynamic t-SNE is an adaptation of t-SNE for sequences of time-dependent datasets. It introduces a controllable trade-off between temporal coherence and projection (embedding) reliability. For more details, please see [2].

If your use of this code results in a publication, please cite (at least) the original paper by Laurens van der Maaten [1].

This implementation is not nearly as computationally efficient as some alternatives (e.g., Barnes-Hut t-SNE [3]).

See the examples directory. Remember to add the directory that contains thesne to the PYTHONPATH.

[1] Van der Maaten, Laurens and Hinton, Geoffrey. Visualizing data using t-SNE. Journal of Machine Learning Research, 2008.

[2] Rauber, Paulo E., Falcão, Alexandre X., Telea, Alexandru C. Visualizing Time-Dependent Data Using Dynamic t-SNE. Proc. EuroVis Short Papers, 2016.

[3] Van der Maaten, Laurens. Accelerating t-SNE using Tree-Based Algorithms. Journal of Machine Learning Research, 2014.