Permutation Search of Tensor Network Structures via Local Sampling (ICML, 2022)

Intro

This repository is the implementation of TN-structure Local Sampling (TNLS) under the ring constraint (arXiv).

Requirements

• Python 3.7.3
  

• Tensorflow 1.13.1

Usage

First you need to start agents with

 CUDA_VISIBLE_DEVICES=0 python agent.py 0

The last 0 stands for the id of the agent. You can spawn multiply agents with each one using one gpu by modifying the visible device id.

Then start the main script by

python TNLS_TR.py data.npz 60

The argvs stands for the name of data, the numbers of samples in one generation. Here we provide a demo of learning the low-dimensional representation of a TR format tensor. The details of the algorithm will be saved in a .log file.

Acknowledgement

• The code is modified based on the TNGA. Thanks for their great efforts.