Low Rank Matrix Completion Benchmarks

This project compares low rank matrix completion algorithms on many different kinds of random and real data matrices. The data and plots are generated by code in this repository (this repo also contains the source of this website). Below, feel free to examine algorithm performance on either random or real data matrices. More options to come soon!

Select a matrix for comparison

Type of Matrix
Random Data Real World Data

Size of (Square) Matrix
1000 10000

Rank of Matrix
2 20 200

Percent of Matrix Shown
5% 10% 20% 40%

Condition Number
2 200 20000

Dataset
Lena Movie Lens

This combination of values is actually below the information theoretic limit for low rank matrix recovery! See this paper for more information. Per that result we require the following condition:
matrix_size^2 * fraction_shown > (2 * matrix_size - rank) * rank