GL2U-II is an open source python/C++ package for computing upper and lower probabilities for non-separately specified, non-binary credal networks. The framework of GL2U-II combines the local specification, the binarization and the loopy 2U inference process, similar to the previous version of GL2U.

However, as a brand new implementation, we introduce in GL2U-II several new binarization methods and an improved L2U implementation, as a result, GL2U-II is outstandingly fast and can completely replace the old GL2U implementation.

Requirements

To run GL2U-II, you need Python version 2.5 or higher, with package NetworkX. Also, you need a C++ compiler which is properly linked to Python.distutil.

Downloads

The latest GL2U-II package (version 0.98), including code, documentation and examples, can be downloaded here.

The benchmarks used for the test can be found here.

A detailed technical description of GL2U-II can be found here.

References

GL2U-II uses the LRS package for extracting vertices of polytopes from interval specifications of credal nets.

The executable files of 2UBayes and LRS can be found in extern folder.