slpKanren
Stochastic Logic Programs (SLP) style probabilistic logic programming in miniKanren, based on Stephen Muggleton's paper, 'Stochastic Logic Programs': http://www.doc.ic.ac.uk/~shm/Papers/slp.pdf
Code by Rebecca Swords and William E. Byrd, based on core miniKanren.
slpKanren extends core miniKanren with one relational operator (condp) and two interface operators (run-prob and run-prob*):
(condp [prob-exp g g* ...] ...)
condp is identitical to conde, except that the first expression in each clause must evaluate to a real number representing the probability associated with that clause. Operationally, condp behaves identically to conde, other than associating a probability with each successful clause. In other words, condp and conde produce the same answers, in the same order; however, condp associates a probility with each answer.
(run-prob n (x) g0 g ...)
(run-prob* (x) g0 g ...)
run-prob and run-prob* are identical to run and run*, except that the probability associated with each answer is also returned.
This implementation also includes two debugging goals, which can be used to examine the substition: print-substo and print-prob-substo.
Example slpKanren program, adapted from the Muggleton paper:
;;; stochastic automaton
(define sa
(lambda (S)
(letrec ([q0 (lambda (S)
(exist (S^)
(condp
[0.4
(== `(a . ,S^) S)
(q0 S^)]
[0.6
(== `(b . ,S^) S)
(q1 S^)])))]
[q1 (lambda (S)
(exist (S^)
(condp
[0.7
(== `(b . ,S^) S)
(q1 S^)]
[0.3
(== `(c . ,S^) S)
(q2 S^)])))]
[q2 (lambda (S) (== '() S))])
(q0 S))))
and associated test cases:
(test "sa-1"
(run-prob 10 (q) (sa q))
'(((b c) . 0.18)
((b b c) . 0.126)
((a b c) . 0.072)
((b b b c) . 0.08819999999999999)
((b b b b c) . 0.06173999999999999)
((a b b c) . 0.0504)
((b b b b b c) . 0.043217999999999986)
((b b b b b b c) . 0.03025259999999999)
((a a b c) . 0.0288)
((a b b b c) . 0.03528)))
(test "sa-2"
(run-prob 1 (q) (== '(a b b c) q) (sa q))
`(((a b b c) . ,(* 0.4 0.6 0.7 0.3))))
All Scheme code tested under Petite Chez Scheme Version 8.4.