{"repo":"LUK4S-B/IsUMap","free":true,"listed":false,"github":"https://github.com/LUK4S-B/IsUMap","clone":"git clone https://github.com/LUK4S-B/IsUMap.git","description":"IsUMap is a tool for manifold learning, dimension reduction and data visualization","language":"Jupyter Notebook","stars":24,"topics":["data-visualization","dimension-reduction","manifold-learning","metric-learning"],"license":null,"category":"analytics","readme_excerpt":"IsUMap IsUMap is a dimension reduction and data visualization tool, that can be viewed as a combination of the manifold learning algorithms UMAP and Isomap. The theoretical basis of IsUMap is explained in the following publications: - Fuzzy simplicial sets and their application to geometric data analysis - IsUMap: Manifold Learning and Data Visualization Leveraging Vietoris-Rips Filtrations Please cite them when using IsUMap. Note that an appendix that contains additional theoretical aspects and numerical experiments is given in the Appendix folder. Usage Right now, we do not yet provide IsUMap as a package. But since it is rather lightweight (except, perhaps, for the pytorch dependency), you can use it simply by cloning the repo, and installing the dependencies in environment.yml . For example, assuming you want to install with conda to a new environment, you can run: After that, you should be able to run python minimal example.py . Other examples are described in the next section below. Examples In minimal example.py , one can choose one of a series of example datasets, simply by uncommenting one of the following lines: The default example is the non-uniform hemisphere, that is supposed to show that IsUMap can successfully uniformize the data distribution. Below we provide a figure that compares some results for low-dimensional manifolds. One can see that IsUMap excels for manifolds with lower dimension because it distorts distances less than UMAP, while being able to unifo","default_branch":null,"files":null,"tree":[],"storefront":"/r/LUK4S-B","claimed":false,"request_supported":{"post":"https://gitbuyer.com/r/LUK4S-B/IsUMap/request-supported","requests":0},"note":"indexed from public GitHub; nothing is for sale on this page. Clone it from GitHub. Paid listings live at /search."}