MultiView Learning Benchmark

A lightweight PyTorch benchmark for multi-view learning and semi-supervised classification, with MLP-, GCN-, and Transformer-based baselines.

Overview

This repository provides a compact and accessible benchmark for multi-view learning. It is designed to help researchers and beginners reproduce baseline experiments, understand common multi-view architectures, and build new methods on top of a consistent experimental pipeline.

Installation

Clone the repository and enter the project directory:

git clone https://github.com/LosparkSayoji/MultiView-Learning-Benchmark.git
cd MultiView-Learning-Benchmark

Install the dependencies:

pip install -r Baseline/requirements.txt
pip install gdown

The main requirements are Python 3, PyTorch 2.1 or later, NumPy, SciPy, scikit-learn, tqdm, and gdown.

Dataset

The multi-view datasets used by this benchmark are available in the following shared Google Drive folder:

Download the multi-view datasets from Google Drive

You can also download the datasets automatically from the repository root:

python Dataset/download_multi-view_data.py --output ./dataset

Configuration

Before running a baseline, open its main.py file and update the dataset list and local data path:

DATASETS_LIST = ["100leaves"]
DATA_ROOT = "/path/to/your/multiview-datasets/"

Other experimental settings can be configured through the corresponding args.py file or command-line arguments.

For example:

python main.py --device cpu --rep_num 5 --num_epochs 200