The t-distributed Stochastic Neighbour Embedding (t-SNE) algorithm has an interesting parameter called 'perplexity'. The effect of perplexity was spectacularly highlighted in the article, How to Use t-SNE Effectively. If I've understood it correctly, perplexity correlates to the number of nearby datapoints, or "neighbours", for every datapoint in lower-dimensional space. Changing this parameter severely impacts the resulting plot.

This was super interesting, so I wanted to play around with it for myself.

Modifying van der Maaten's python implementation, and using MatPlotLib for the visualisation, I created animations to have a look at how perplexity affects simplistic datasets.

More to come.