PySDD 0.2.2

Sentential Decision Diagrams

Python wrapper package to interactively use Sententical Decision Diagrams (SDD).

This is a beta version, not yet all functions in sddapi.h are available.

Full documentation available on http://pysdd.readthedocs.io.

Dependencies

• Python >=3.6

• Cython

• SDD package >=2.0: http://reasoning.cs.ucla.edu/sdd/

Installation

$ pip install git+https://github.com/wannesm/PySDD.git#egg=PySDD

Compilation

Notice: This wrapper requires some small changes to the SDD package. The changed files are already included in this repository. Do not overwrite them with the original files.

• Download the SDD package from http://reasoning.cs.ucla.edu/sdd/.

• Install the SDD package in the PySDD package in directories pysdd/lib/sdd-2.0 and pysdd/lib/sddlib-2.0 without overwriting the already available files.

• Run python3 setup.py build_ext --inplace or make build to compile the library in the current directory. If you want to install the library such that the library is available for your local installation or in your virtual environment, use python3 setup.py install.

For some Linux platforms, it might be necessary to recompile the libsdd-2.0 code with the gcc option -fPIC and replace the pysdd/lib/sdd-2.0/lib/Linux/libsdd.a library with your newly compiled version.

Python package

The wrapper can be used as a Python package and allows for interactive use.

The following example builds an SDD for the formula a∧b ∨ b∧c ∨ c∧d.

from pysdd.sdd import SddManager, Vtree, WmcManager
vtree = Vtree(var_count=4, var_order=[2,1,4,3], vtree_type="balanced")
sdd = SddManager.from_vtree(vtree)
a, b, c, d = [sdd.literal(i) for i in range(1, 5)]

# Build SDD for formula
formula = (a & b) | (b & c) | (c & d)

# Model Counting
wmc = formula.wmc(log_mode=False)
print(f"Model Count: {wmc.propagate()}")
wmc.set_literal_weight(a, 0.5)
print(f"Weighted Model Count: {wmc.propagate()}")

# Visualize SDD and Vtree
with open("output/sdd.dot", "w") as out:
    print(formula.dot(), file=out)
with open("output/vtree.dot", "w") as out:
    print(vtree.dot(), file=out)

The SDD and Vtree are visualized using Graphviz DOT:

More examples are available in the examples directory. An interactive Jupyter notebook is available in notebooks/examples.ipynb

Command Line Interface

A Python CLI application is installed if you use pip, pysdd. Or it can be used directly from the source directory where it is called pysdd-cli.py. This script mimicks the original sdd binary and adds additional features (e.g. weighted model counting)

$ pysdd -h
$ ./pysdd-cli.py -h
usage: pysdd-cli.py [-h] [-c FILE | -d FILE | -s FILE] [-v FILE] [-W FILE]
                [-V FILE] [-R FILE] [-S FILE] [-m] [-t TYPE] [-r K] [-q]
                [-p] [--log_mode]

Sentential Decision Diagram, Compiler

optional arguments:
  -h, --help  show this help message and exit
  -c FILE     set input CNF file
  -d FILE     set input DNF file
  -s FILE     set input SDD file
  -v FILE     set input VTREE file
  -W FILE     set output VTREE file
  -V FILE     set output VTREE (dot) file
  -R FILE     set output SDD file
  -S FILE     set output SDD (dot) file
  -m          minimize the cardinality of compiled sdd
  -t TYPE     set initial vtree type (left/right/vertical/balanced/random)
  -r K        if K>0: invoke vtree search every K clauses. If K=0: disable
              vtree search. By default (no -r option), dynamic vtree search is
              enabled
  -q          perform post-compilation vtree search
  -p          verbose output
  --log_mode  weights in log

Weighted Model Counting is performed if the NNF file containts a line
formatted as follows: "c weights PW_1 NW_1 ... PW_n NW_n".

Memory management

Python’s memory management is not used for the internal datastructures. Use the SDD library’s garbage collection commands (e.g. ref, deref) to perform memory management.

References

This package is inspired by the SDD wrapper used in the probabilistic programming language ProbLog.

References:

• Wannes Meert, PySDD, in Recent Trends in Knowledge Compilation, Report from Dagstuhl Seminar 17381, Sep 2017. Eds. A. Darwiche, P. Marquis, D. Suciu, S. Szeider.

Other languages:

• C: http://reasoning.cs.ucla.edu/sdd/

• Java: https://github.com/jessa/JSDD

Contact

• Wannes Meert, KU Leuven, https://people.cs.kuleuven.be/wannes.meert

• Arthur Choi, UCLA, http://web.cs.ucla.edu/~aychoi/

License

Python SDD wrapper:

Copyright 2017-2018, KU Leuven and Regents of the University of California. Licensed under the Apache License, Version 2.0.

SDD package:

Copyright 2013-2018, Regents of the University of California Licensed under the Apache License, Version 2.0.