This repository is the implementation of TnALE under the ring constraint.

• Python 3.7.3
  

• Tensorflow 1.13.1

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 multiple agents with each one using one GPU by modifying the visible device id.

Then start the main script by

 python TNALE_TR.py 'data.npz' 2 1 1 2

The argvs stand for the name of data, the rank-related radius in the initial phase, the rank-related radius in the main phase, the switch that decides whether or not to include the initial phase and the $L_{0}$ in the paper, respectively. Here we provide a demo of learning the low-dimensional TR format representation of a tensor. The running details will be saved in a .log file.

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