Nonparametric Decomposition of Sparse Tensors
This is the implementation of the NEST-1 and NEST-2 sparse tensor decomposition algorithms from the paper Nonparametric Decomposition of Sparse Tensors, by Conor Tillinghast and Shandian Zhe @ The Thirty-eighth International Conference on Machine Learning (ICML), 2021.
MIT license
All code were tested under python 3.6 and TensorFlow 1.12.0. We recommend creating a virtual environment and pip installing the requirements.txt file.
To run the algorithms run either main/run_nest1_trial.py or main/run_nest2_trial.py from the command line. Each requires the path to the training data, the path to the test data and the rank of the decomposition. The batch size, learning rate and number of epochs can be specified as optional arguments. For example:
ex) python main/run_nest2_trial.py "data/alog/train-fold-1.txt" "data/alog/test-fold-1.txt" 3 --nepoch 700 --lr 0.001
If you use our code please cite our paper
@article{ctill2021nest,
title={Nonparametric Decomposition of Sparse Tensors},
author={Tillinghast, Conor and Zhe, Shandian},
journal={The Thirty-eighth International Conference on Machine Learning},
year={2021}}