Neural Additive Tensor Decomposition for Sparse Tensors

This is a PyTorch implementation of "Neural Additive Tensor Decomposition for Sparse Tensors". This paper proposed NeAT, a tensor decomposition methods discover non-linear latent patterns in tensors in an interpretable way.

Prerequisites

Before you begin using this code, make sure you have the following libraries installed:

• Python 3.9
• PyTorch
• NumPy
• DotMap
• TensorLy
• torchmetrics

Usage

To run the demo script, simply execute the demo.sh script or run the demo.ipynb

Directory structure

• configs/ # Configuration files for datasets.
• dataset/ # Contains datasets for experimentation.
  • dblp # Dataset for an inductive setting.
  • trans_dblp # Dataset for a transductive setting.
  • ml
  • yelp
  • foursquare_nyc
  • foursquare_tky
  • yahoo_music
• src/
  • main.py # Running NeAT.
  • read.py # Reading datasets.
  • model.py # NeAT model.
  • train.py # Training NeAT.
  • utils.py # Saving models.
  • metrics.py # Evaluation metrics.
• run.sh # script for executing the main.py
• README.md # This documentation file.