A tool to verify algorithmic fairness properties, as seen in the OOPSLA 2017 paper Probabilistic Verification of Program Fairness by Aws Albarghouthi, Loris D'Antoni, Samuel Drews, and Aditya Nori.
FairSquare runs on Linux and uses Python 3.
Additionally, it uses the following dependencies:
z3,
the SciPy stack,
and the python packages codegen and asteval
(easily obtained using pip:
pip install --user codegen asteval)
By default, FairSqaure uses
redlog
for quantifier elimination, which must be downloaded separately:
FairSquare expects the downloaded files to be placed in src/tools/
(specifically, it runs src/tools/reduce,
which relies on the existence of src/tools/reduce.img).
Alternatively, FairSquare can use a DNF-based quantifier elimination
implemented with z3 by using the -z flag when running the tool.
Once the dependencies are installed, navigate to the src directory.
Basic usage of the tool involves a command like
python fairProve.py -f ex.fr -mi -mf
-f ex.fr specifies that ex.fr contains the fairness verification problem to be analyzed
(in fact, src/ex.fr contains the illustrative example from the OOPSLA paper),
and -mi -mf are optimizations that should always be used.
A full list of command-line flags is available from
python fairProve.py -h.
You can reproduce our published benchmarks by navigating
to the oopsla directory (at the top level).
The benchmarks are divided, based on their fairness postconditions,
into oopsla/qual and oopsla/noqual.
The oopsla/run.sh script can be used to run batches of benchmarks;
see oopsla/README for more information.
- Samuel Drews (primary contact)
- David Merrell
- Aws Albarghouthi
- Loris D'Antoni
- Aditya Nori