/fealm

Primary LanguagePythonBSD 3-Clause "New" or "Revised" LicenseBSD-3-Clause

Feature Learning for Nonlinear Dimensionality Reduction toward Maximal Extraction of Hidden Patterns

About

  • The implementation of FEALM, FEALM-UMAP, and NSD from: Feature Learning for Nonlinear Dimensionality Reduction toward Maximal Extraction of Hidden Patterns. Takanori Fujiwara, Yun-Hsin Kuo, Anders Ynnerman, and Kwan-Liu Ma. In Proc. PacificVis, 2023. arXiv Preprint)

  • Demonstration video: https://takanori-fujiwara.github.io/s/fealm/index.html

  • Features

    • Feature learning framework, FEALM, for nonliner dimensionality reduction methods.

    • Exemplifying method of FEALM for UMAP (FEALM-UMAP)

    • Various graph dissimilarity measures, including neighbor-disimilarity measure (NSD)

    • Implementation of "Optimal Sets of Projections of High-Dimensional Data" [Lehmann and Theisel, 2016]

    • Generation code of the 2-spheres and 3-class attribute data, which is designed to illustrate nonlinear dimensionality reduction’s sensitiveness to trivial disturbance to manifolds

Content

  • fealm: FEALM framework and the exemplifying method for UMAP.
  • motivating_examples: Scripts used for Motivating Examples.
  • computational_evaluations: Scripts used for Computational Evaluations.
  • case_studies: Scripts related to Case Studies.

Setup

Requirements

  • Python3

  • Note: Tested on macOS Sonoma, Ubuntu 22.0.4 LTS, and Windows 10.

Setup

  • Install FEALM

    • Download/Clone this repository

    • If using Python3.12 (otherwise, skip this step), need to edit and directly install pymanopt as follows because pymanopt installer only supports Python3.11 or older.

      • Download and move to the latest pymanopt repo: https://github.com/pymanopt/pymanopt.

      • Edit pyproject.toml

        • Line 2: requires = ["pip>=22.3.1", "setuptools>=65.6.3"]

        • Line 40: "scipy>=1.0",

      • Create "_version.py" in "src/pymanopt/" and write down a following line:

        • __version__ = "2.2.0"

      • Install pymanopt

        • pip3 install .

    • Move to the downloaded fealm repository, then:

      pip3 install .

  • You can test with sample.py. To run, sample.py you need to install matplotlib.

    pip3 install matplotlib

Usage

  • Examples can be found in "sample.py" and "case_studies".

  • API documentation can be found in "fealm/fealm.py".

  • Currently, due to the conflict between UMAP and Pathos libraries, a specific loading order of modules is required (see sample.py).

How to cite

T. Fujiwara, Y.-H. Kuo, A. Ynnerman, and K.-L. Ma, "Feature Learning for Nonlinear Dimensionality Reduction toward Maximal Extraction of Hidden Patterns." In Proc. PacificVis, 2023. (https://arxiv.org/abs/2206.13891)