pyapproxmc 4.3.2

Bindings to ApproxMC, an approximate model counter

pyapproxmc: Python bindings to ApproxMC

Python bindings to ApproxMC, an approximate model counter with PAC (Probably Approximately Correct) guarantees. ApproxMC counts the number of satisfying assignments of CNF formulas.

Installing

pip install pyapproxmc

Building from source

# Install build tools
pip install scikit-build-core build

# Clone the repository
git clone --recurse-submodules https://github.com/meelgroup/approxmc
cd approxmc

# Build and install into a virtual environment
python -m venv .venv && source .venv/bin/activate
pip install .

The build system uses scikit-build-core and CMake. All C++ dependencies (CryptoMiniSat, Arjun, and their sub-dependencies) are fetched and built automatically by CMake during the pip install step. The only system library required is GMP:

• Linux: sudo apt-get install libgmp-dev (or dnf install gmp-devel)
• macOS: brew install gmp

Usage

import pyapproxmc

c = pyapproxmc.Counter()
c.add_clause([1, 2, 3])
c.add_clause([3, 20])
count = c.count()
print("Approximate count is: %d*2**%d" % (count[0], count[1]))

count() may only be called once per Counter instance; create a new Counter if you need to count a different formula.

The above prints Approximate count is: 11*2**16. Since the largest variable in the clauses is 20, the formula has at most 2**20 models; the two clauses restrict this to approximately 11*2**16 ≈ 720896 models.

Counting over a projection set

Pass a projection set (sampling set) to count() to count models projected onto a subset of variables:

import pyapproxmc

c = pyapproxmc.Counter()
c.add_clause([1, 2, 3])
c.add_clause([3, 20])
count = c.count(range(1, 10))
print("Approximate count is: %d*2**%d" % (count[0], count[1]))

This prints Approximate count is: 7*2**6, the approximate count projected over variables 1–9.

Adding clauses from arrays

For performance-critical code, multiple clauses can be added in one call via add_clauses() with a flat, zero-terminated array.array. This uses the buffer protocol and avoids per-element Python overhead:

import pyapproxmc
from array import array

c = pyapproxmc.Counter()
# Clauses [1,2,3] and [3,20], each terminated by 0
c.add_clauses(array('i', [1, 2, 3, 0, 3, 20, 0]))
count = c.count()

Note: add_clause() (singular) also accepts any iterable including arrays, but iterates element by element and does not use the buffer protocol.

Counter constructor parameters

Parameter	Default	Description
seed	1	Random seed for reproducibility
verbosity	0	Output verbosity (0 = silent)
epsilon	0.8	Tolerance: how approximate the count is
delta	0.2	Confidence: probability the count is within tolerance

Example:

c = pyapproxmc.Counter(seed=42, epsilon=0.5, delta=0.1)

Version

import pyapproxmc
print(pyapproxmc.__version__)