This repository describes (in code and rationale) a methodology that we explained in
Flow-based Algorithms for Improving Clusters: A Unifying Framework,
Software, and Performance. Kimon Fountoulakis, Meng Liu,
David F. Gleich, Michael W. Mahoney. https://arxiv.org/abs/2004.09608
The idea is that we can generalize various spectral or eigenvector-based visualization techniques by exploiting a relationship between them and random samples of the SVD. This is described above in more detail. Also, see the video (link coming soon).
To understand this repository, please read
- `live-notes.txt` *A set of overview notes we took as we proceeded*
- `live-spectral-vis.jl` *This describes the spectral scenario*
- `live-flow-intro.jl` *This describes the nonlinear process*
- `live-flow-problem.jl` *This describes the issue we try and fix*
- `live-sampled-vis.jl` **This describes our idea!**
- `live-fashion-mnist.jl` *Another example of our idea*
There are two graphs used in this experiment. They are listed in the data section.