AAAI 2026-KNNDA

This project contains the code and data of the accepted paper for AAAI 2026: KNNDA: A New Perspective of Alignment Recovery for Partially View-Aligned Clustering

Requirements

pytorch==2.0.0

numpy==1.24.3

scikit-learn==1.3.0

munkres==1.1.4

Configuration

The parameters and the training options (including the epochs and aligned proportions) are defined in the corresponding .yaml file, e.g. Scene15.yaml.

Datasets

All the datasets are placed in "datasets" folder.

Usage

The code and configs are set by default for the two-view experiment. Run the following code to obtain the clustering results on the Scene15 dataset.

python run.py --data 0

If you want to run the three-view experiment on 100Leaves and BBCsports3view datasets, please change the value of the --data argument to the corresponding value which can be checked in run.py. Then modify the load_alldata function to load corresponding views. You also need to uncomment the code in the forward function of corresponding model class in models.py.