GLOP

This article is about the linear programming solver. For the multicast address allocation method, see Multicast address § GLOP addressing.

GLOP (the Google Linear Optimization Package) is Google's open source linear programming solver, created by Google's Operations Research Team. It is written in C++ and was released to the public as part of Google's OR-Tools software suite in 2014.^[1]

GLOP uses a revised primal-dual simplex algorithm optimized for sparse matrices. It uses Markowitz pivoting to reduce matrix fill-in, steepest-edge pricing to avoid degenerate pivots, and an LU decomposition tailored for sparse matrices.

Inside Google, GLOP is used to stabilize YouTube videos^[2] and outside Google, it has been used to perform fast linear relaxations for reinforcement learning.^[3]

References

1. "Sudoku, Linear Optimization, and the Ten Cent Diet".
2. "Sudoku, Linear Optimization, and the Ten Cent Diet".
3. "A structured prediction approach for generalization in cooperative multi-agent reinforcement learning".

External links

• GLOP home page
• GLOP source code