Approximate UMAP

The approximate UMAP algorithm

Approximate UMAP is a UMAP variant made for projecting novel data at a rapid pace suitable for online, real-time introspection. Approximate UMAP utilizes a simplified approach to transform novel data, circumventing the time-costly graph optimization otherwise performed by standard UMAP. Approximate UMAP does not use an alternative method for fitting a model, in this regard, it is identical to standard UMAP. See the below-mentioned paper for more details. Approximate umap is available as a python package that can be installed via pip:

pip install approx-umap

ONEP - Online Neurodata and Embedding Projector

ONEP is an extension module for the Dareplane framework that supports online, real-time visualization of data such as neural network embeddings. It utilizes a web interface to display a continuously updating figure, allowing a user to track trends in the input data. ONEP may be used with various projection methods, both existing and custom ones. It already supports the following methods: UMAP, approximate UMAP, and CEBRA.

Michael Tangermann
Michael Tangermann
Head of the Lab, Associate Professor, PI