flamapy-sharpsat 2.6.0.dev12

flamapy-sharpsat is a plugin to flamapy providing approximate model counting and almost-uniform sampling

flamapy-sharpsat

flamapy-sharpsat is a flamapy plugin that adds scalable approximate model counting and almost-uniform sampling of feature-model configurations, on top of the CNF produced by the SAT metamodel.

It complements the existing backends:

• The SAT (flamapy-sat) and Z3 backends count configurations by full enumeration, which does not scale; the BDD backend counts exactly but the diagram can blow up on large models.
• flamapy-sharpsat uses ApproxMC for an (epsilon, delta) approximate count and UniGen for almost-uniform sampling — both hashing-based tools from the CryptoMiniSat family that scale to far larger models.

Installation

pip install flamapy-sharpsat

This pulls in pyunigen (which bundles ApproxMC), and the flamapy core/FM/SAT plugins.

Operations

Operation	Class
Approximate number of configurations	SharpSATConfigurationsNumber
Almost-uniform sampling	SharpSATSampling

Usage

from flamapy.metamodels.fm_metamodel.transformations import UVLReader
from flamapy.metamodels.sharpsat_metamodel.transformations import FmToSharpSAT
from flamapy.metamodels.sharpsat_metamodel.operations import (
    SharpSATConfigurationsNumber,
    SharpSATSampling,
)

fm = UVLReader('model.uvl').transform()
model = FmToSharpSAT(fm).transform()

count = SharpSATConfigurationsNumber().execute(model).get_result()

sampler = SharpSATSampling()
sampler.set_sample_size(10)
sample = sampler.execute(model).get_sample()

Counting and sampling are projected onto the feature variables, so auxiliary variables (such as the ones introduced by the optional Tseytin CNF encoding, available through FmToSharpSAT(fm, cnf_method='tseytin')) never distort the results.

License

GPL-3.0-or-later.