Interpreting t-SNE Plots
Now that we are able to use t-distributed SNE to visualize high-dimensional data, it is important to understand the limitations of such plots and what aspects are important in interpreting and generating them. In this section of the chapter, we will highlight some of the important features of t-SNE and demonstrate how care should be taken when using the visualization technique.
Perplexity
As described in the introduction to t-SNE, the perplexity values specify the number of nearest neighbors to be used in computing the conditional probability. The selection of this value can make a significant difference to the end result; with a low value of perplexity, local variations in the data dominate because a small number of samples are used in the calculation. Conversely, a large value of perplexity considers more global variations as many more samples are used in the calculation. Typically, it is worth trying a range of different values to investigate the effect of perplexity...
