richardxie1119 / tensor_completion

Low-rank tensor completion algorithm - HaLRTC.

Tensor Completion

This post introduces a Julia implementation for an efficient low-rank tensor completion algorithm - HaLRTC.

1. Fundamentals of tensor completion

• Tensor unfolding: convert a higher-order tensor into a matrix along specific direction. (see ten2mat.jl)

• Matrix folding: convert a matrix into a higher-order tensor with specific direction and size. (see mat2ten.jl)

2. Tensor completion

• Data set: publicly available urban traffic speed data set in Guangzhou, China can be found from https://doi.org/10.5281/zenodo.1205228. In this evaluation, we have organized the original data into a tensor, and there are two .mat files, including a data tensor tensor.mat and a tensor random_tensor.mat with random numbers.

• Inference: see paper https://doi.org/10.1109/TPAMI.2012.39.

• Evaluation: you can easily accomplish a tensor completion task for missing data imputation by running main_test.jl.

If you have any questions, feel free to contact me with following email: [e-mail].