TedNet: A Pytorch Toolkit for Tensor Decomposition Networks

tednet is a toolkit for tensor decomposition networks. Tensor decomposition networks are neural networks whose layers are decomposed by tensor decomposition, including CANDECOMP/PARAFAC, Tucker2, Tensor Train, Tensor Ring and so on. For a convenience to do research on it, tednet provides excellent tools to deal with tensorial networks.

Now, tednet is easy to be installed by pip:

pip install tednet

More information could be found in Document.

Quick Start

Operation

There are some operations supported in tednet, and it is convinient to use them. First, import it:

import tednet as tdt

Create matrix whose diagonal elements are ones:

diag_matrix = tdt.eye(5, 5)

A way to transfer the Pytorch tensor into numpy array:

diag_matrix = tdt.to_numpy(diag_matrix)

Similarly, the numpy array can be taken into Pytorch tensor by:

diag_matrix = tdt.to_tensor(diag_matrix)

Tensor Decomposition Networks (Tensor Ring for Sample)

To use tensor ring decomposition models, simply calling the tensor ring module is enough.

import tednet.tnn.tensor_ring as tr

# Define a TR-LeNet5
model = tr.TRLeNet5(10, [6, 6, 6, 6])

Citing

If you use tednet in an academic work, we will appreciate you for citing our paper with:

@article{DBLP:journals/ijon/PanWX22,
  author    = {Yu Pan and
               Maolin Wang and
               Zenglin Xu},
  title     = {TedNet: {A} Pytorch toolkit for tensor decomposition networks},
  journal   = {Neurocomputing},
  volume    = {469},
  pages     = {234--238},
  year      = {2022}
}