predpca-py 0.1.2

PredPCA: Predictive Principal Component Analysis

Predictive principal component analysis (PredPCA) - Python Implementation

Overview

Predictive principal component analysis (PredPCA) is an analytically solvable unsupervised learning scheme that extracts the most informative components for predicting future inputs. It is a convex optimization and can find the analytical expressions of optimal weight matrices and encoding dimensionality that provide the global minimum of the squared test prediction error.

This repository involves Python scripts for PredPCA, which is a Python implementation of the original MATLAB code repository.

Paper

"Dimensionality reduction to maximize prediction generalization capability" Takuya Isomura, Taro Toyoizumi Nature Machine Intelligence 3, 434–446 (2021). https://doi.org/10.1038/s42256-021-00306-1 Preprint version is available at https://arxiv.org/abs/2003.00470

Copyright

Original MATLAB implementation: Copyright (C) 2020 Takuya Isomura (RIKEN Center for Brain Science) Python implementation: Copyright (C) 2024 Araya Inc.

Example Notebooks

The following Colab notebooks provide explanations and usage examples of PredPCA:

MNIST Handwritten digits sequence (Fig.2)

ALOI Rotating objects sequence (Fig.3)

Nonlinear Nonlinear time series (Suppl. Fig.1)

Contents

predpca/
├── predpca.py  # Core functions for PredPCA
├── aloi/       # Demo with the ALOI rotating 3D object video dataset (Fig.3 in the paper)
├── bdd100k/    # Demo with the BDD100K naturalistic driving video dataset (Fig.4 in the paper)
├── mnist/      # Demo with the MNIST handwritten digit image dataset (Fig.2 in the paper)
└── nonlinear/  # Demo with nonlinear time series data (Supplementary Fig.1 in the paper)

See each directory for more details.

Installation

It is recommended to use a virtual environment for installation. This project has been tested with Python 3.10.

1. Create and activate a virtual environment (venv for example)

   python -m venv venv
   source venv/bin/activate  # On Windows, use venv\Scripts\activate.ps1 or activate.bat
   

2. Install dependencies

• FFmpeg
  • Install from the official download links or package managers, e.g.
    • Windows: winget install ffmpeg
    • macOS: brew install ffmpeg
    • Ubuntu: sudo apt install ffmpeg
• (optional) PyTorch and Torchvision
  • These packages are required if you want to run the BDD100K demo (predpca/bdd100k/) or model comparison experiments (predpca/*/main_compare_models.py).
  • Install from the official website.
• (optional) deeptime
  • This package is required if you want to run the model comparison experiments.
  • pip install deeptime

3. Install the package

• Option 1: Install from PyPI

  pip install predpca-py
  

• Option 2: Install from source (for developers)

  git clone https://github.com/takuyaisomura/predpca-py.git
  cd predpca-py
  pip install -e .
  

License

This project is covered under the GNU General Public License v3.0.