/multidr

MulTiDR: A Visual Analytics Framework for Reviewing Multivariate Time-Series Data with Dimensionality Reduction

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

MulTiDR: A Visual Analytics Framework for Reviewing Multivariate Time-Series Data with Dimensionality Reduction

About

  • MulTiDR is from: Fujiwara et al., "A Visual Analytics Framework for ReviewingMultivariate Time-Series Data with Dimensionality Reduction." 2020.

  • Implementation of MulTiDR back-end

  • Two-step DR (TDR): Framework of dimensionality reduction for multivariate time-series data. TDR produces a low-dimensional representation from a third-order tensor.

  • Contrastive learning with sign adjustment of feature contributions.

  • Implementation of MulTiDR front-end (The source code will tentatively be released in the future).

  • Demonstration video of a system using MulTiDR: https://takanori-fujiwara.github.io/s/multidr/

Requirements

  • Python3
  • Note: Tested on macOS Catalina and Ubuntu 20.0.4 LTS.

Setup

  • Install with pip3. Move to the directory of this repository. Then,

    pip3 install .

  • If you want to use contrastive learning with a default setting (i.e., use of ccPCA), install ccPCA from: https://github.com/takanori-fujiwara/ccpca

Usage

  • Import installed modules from python (e.g., from multidr.tdr import TDR). See sample.py for examples.
  • For detailed documentations, please see doc/index.html or directly see comments in multidr/tdr.py and multidr/cl.py.

How to Cite

Please, cite:
Takanori Fujiwara, Shilpika, Naohisa Sakamoto, Jorji Nonaka, Keiji Yamamoto, and Kwan-Liu Ma, "A Visual Analytics Framework for ReviewingMultivariate Time-Series Data with Dimensionality Reduction". arXiv:2008.01645, 2020.