t-Distributed Stochastic Neighbor Embedding, or t-SNE, like other data reduction techniques as PCA and MDS, creates new smaller set of variables out of a large number of variables, retaining much of the information of the original data-set, but has a different approach to achieving this. What does this method add to these others? In my experience so far, it serves my gread to give me most predictability and the serves my desparation in the exploratory ‘get me a gist of the data’ phase. In other words: it improves the goodies of PCA and does not too much damage to representing a dataset in comparison to MDS.

So how does it work? Some of the workings of this approach are in the name of method. The method is al about neighbouring points and embedding these neighboring points in local neighborhoods fitting them to a lower dimensionsional space. The t-Distributiony and stochasticacy part need further explanation. At least: that’s what my first thoughts were about these words… The result is clustering of the observations around

In this R tutorial I’ll be using the Barnes-Hut implementation of t-SNE from the Rtsne library. The biggest reason for this is it’s quick processing and little else, since I have little patience for my computer and too little time to read endlessly.