This is our implementation for the paper Probabilistic Neural Kernel Tensor Decomposition, by Conor Tillinghast, Shikai Fang, Kai Zheng, and Shandian Zhe @ IEEE International Conference on Data Mining (ICDM), 2020.
MIT license
All code were tested under python 3.6 and TensorFlow 1.14.0.
The main.py file takes five arguments, which are the path of input training file, the path of testing file, decomposition rank, and tensor dimensions. Batch size, the number of epochs and learning rate are option arguments. The following would give a rank 3 decomposition of a (200,100,200) tensor with learning rate .01, batch size 128 for 50 epochs
ex) python main.py -tr train.txt -te test-fold.txt -r 3 -dim 200 100 200 -lr .01 -ne 50 -bs 128
The default settings for a batch size of 256, a learning rate of .001 and 100 epochs.
We also include our implementation of GPTF
If you use our code please cite our paper
@article{ctill2020pond,
title={Probabilistic Neural Kernel Tensor Decomposition},
author={Tillinghast, Conor and Fang, Shikai and Zheng, Kai and Zhe, Shandian},
journal={IEEE International Conference on Data Mining},
year={2020}}