Below are some UMAP projections of UEA/UCR time-series classification datasets.
In each dataset, each time series is represented in the high-dimensional feature space of hctsa features. In each plot, we show the data projected into a two-dimensional UMAP space. Each point in the space is a time series, and we have coloured each point according to its assigned label. In many cases, the unsupervised UMAP projection of the hctsa feature space captures the labeled structure of the data.
Only a selection of the most interesting-looking projections (according to our eyes) are shown here. Oh, did you want to have a play with all the data? It's available on figshare.
Thanks to Carl Lubba for doing all the computation shown here!
