https://www.youtube.com/watch?v=NEaUSP4YerM&t=1s&ab_channel=StatQuestwithJoshStarmer
UMAP and tsne manage to reduce dimensionality while maintaining the topography of the data. As a side effect they are much slower than PCA or SVD. To speed it up you can use NVIDIA tool called RAPIDS (more in Kaggle Book p. 338 43%).
The openTSNE is a modular Python implementation of t-Distributed Stochasitc Neighbor Embedding (tsne), a popular dimensionality-reduction algorithm for visualizing high-dimensional data sets. openTSNE incorporates the latest improvements to the t-SNE algorithm, including the ability to add new data points to existing embeddings, massive speed improvements, enabling tsne to scale to millions of data points and various tricks to improve global alignment of the resulting visualizations.
Both, UMAP and tsne have to be used carefully, because it is easy to spot clusters and patterns where there are none. The following articles offer advice to use both techniques properly: