Run probabilistic programs¶

ProbLog extends logic programming with probabilities: clauses may be annotated as P::Clause, meaning that the clause holds with probability P, independently of the others. A ProbLog solver computes, for each solution of a query, the probability that it holds. This guide shows how to run ProbLog programs with 2PPy.

Parse probabilistic clauses¶

The :: operator is not part of standard Prolog: parse ProbLog programs with tuprolog.solve.problog.PROBLOG_OPERATORS:

>>> from tuprolog.solve.problog import PROBLOG_OPERATORS, problog_solver
>>> from tuprolog.theory.parsing import parse_theory
>>> graph = parse_theory(
...     """
...     0.6::edge(a, b).
...     0.7::edge(b, c).
...     0.4::edge(a, c).
...     path(X, Y) :- edge(X, Y).
...     path(X, Y) :- edge(X, Z), path(Z, Y).
... """,
...     PROBLOG_OPERATORS,
... )

Here each edge of a graph exists with some probability, while path/2 is an ordinary rule.

Compute probabilities¶

Create a solver with problog_solver (it takes the same arguments as prolog_solver), and solve queries with the probabilistic=True option. tuprolog.solve.plp.probability reads the probability of each solution:

>>> from tuprolog.solve import solve_options
>>> from tuprolog.solve.plp import probability
>>> solver = problog_solver(static_kb=graph)
>>> for solution in solver.solve("path(a, X)", solve_options(probabilistic=True)):
...     print(solution.solved_query, round(probability(solution), 3))
path(a, b) 0.6
path(a, c) 0.652

path(a, c) holds if the direct edge exists, or if both a-b and b-c do: 0.4 + 0.6 × 0.7 − 0.4 × 0.6 × 0.7 = 0.652. ProbLog solutions are distinct: each answer appears once, with the probability of all its proofs combined.

Note

Use solve, not solve_once: the first solution of solve_once accounts only for the first proof (here, 0.4 for path(a, c)).

Without probabilistic=True, a ProbLog solver behaves as a Prolog one, enumerating proofs:

>>> list(solver.query("path(a, X)"))
[{'X': 'b'}, {'X': 'c'}, {'X': 'c'}]

Probabilistic facts with alternative explanations¶

A goal provable in several ways is more likely than each way alone:

>>> weather = parse_theory(
...     """
...     0.3::rain.
...     0.4::sprinkler.
...     wet :- rain.
...     wet :- sprinkler.
... """,
...     PROBLOG_OPERATORS,
... )
>>> solver = problog_solver(static_kb=weather)
>>> [round(probability(s), 2) for s in solver.solve("wet", solve_options(probabilistic=True))]
[0.58]

Inspect the decision diagrams¶

Probabilistic solutions carry the binary decision diagram (BDD) ProbLog built to compute their probability, available via tuprolog.solve.plp.binary_decision_diagram.

Use the ProbLog REPL¶

python -m tuprolog.repl.problog opens an interactive ProbLog shell, printing the probability of each solution: see Use the REPL.