A system for defeasible reasoning over weighted knowledge bases.
- Python 3.10+
Via pip:
$ pip install valphiAlternative, clone the repository and use poetry:
$ poetry installRun the CLI with either
$ python -m valphior
(valphi) $ ./valphi_cli.pyfollowed by the required arguments.
To find all solutions of a multi-layer network, for example the one in examples/kbmonk1.network, run
(valphi) $ ./valphi_cli.py --network-topology examples/kbmonk1.network --weight-constraints --ordered solveTo answer a query use
(valphi) $ ./valphi_cli.py --network-topology examples/kbmonk1.network --weight-constraints --ordered query --query-filename examples/kbmonk1-1.queryA description of the available options is given by
(valphi) $ ./valphi_cli.py --help
(valphi) $ ./valphi_cli.py solve --help
(valphi) $ ./valphi_cli.py query --helpThe input layer is implicit.
Each one of the other layers is encoded by a sequence of lines terminated by a line containing #.
Each line encodes a node of the layer, and consists of a space-separated list of weights.
The first weight is the bias of the node, the other weights are the influence of nodes in the previous layer.
Layers are indexed starting by 1, and can be marked as crisp (binary evaluation) by adding a line
crisp INDEX
Exactly-one constraints on the nodes of the input layer (layer 1) can be specified by adding lines of the form
=1 NODE-INDEX NODE-INDEX ...
where each NODE-INDEX is the index of a node in the input layer, again indexed starting by 1.
The first line is
#graph
The other lines have three integers encoding a weighted link of the graph.
valphi can solve the MAXSAT problem for CNF formulas encoded in the DIMACS format, but don't expect it to be competitive with MAXSAT solvers.
In fact, in this context the solving approach of valphi is not optimized, and a simple reduction (or compilation) of the problem is adopted.