How to shuffle dynamic facts in Prolog? - prolog

Question

I have several dynamic facts in Prolog and I want to shuffle them (reorder in random order). Is there any way in Prolog how to do this?
:- dynamic max/3.
max(1,2,3).
max(1,5,6).
max(3,4,5).
max(2,2,5).
Possible random order:
max(2,2,5).
max(1,2,3).
max(3,4,5).
max(1,5,6).

Answer 1

As you mention that you're using SWI-Prolog, a possible solution is to use its nth_clause/3 and clause/3 built-in predicates. The idea is to access the predicate using a proxy predicate, ramdom_max/3 in this case. I'm also assuming that you only have facts.
:- use_module(library(lists)).
:- use_module(library(random)).
ramdom_max(A, B, C) :-
predicate_property(max(_,_,_), number_of_clauses(N)),
numlist(1, N, List),
random_permutation(List, Permutation),
member(Index, Permutation),
nth_clause(max(_,_,_), Index, Ref),
clause(max(A,B,C), _, Ref).
Sample call:
?- ramdom_max(A, B, C).
A = 1,
B = 2,
C = 3 ;
A = 3,
B = 4,
C = 5 ;
A = 1,
B = 5,
C = 6 ;
A = B, B = 2,
C = 5.
Each call to the ramdom_max/3 predicate will give you a different clause random order but still enumerating all the clauses on backtracking.
This, however, is a relatively computationally costly solution. But as max/3 is a dynamic predicate the first goals in the body of the ramdom_max /3 clause cannot be optimized to run only once. Let's check the number of inferences:
% autoload the time/1 library predicate:
?- time(true).
% 3 inferences, 0.000 CPU in 0.000 seconds (60% CPU, 333333 Lips)
true.
?- time(ramdom_max(A, B, C)).
% 42 inferences, 0.000 CPU in 0.000 seconds (85% CPU, 913043 Lips)
A = 3,
B = 4,
C = 5 ;
% 6 inferences, 0.000 CPU in 0.000 seconds (69% CPU, 272727 Lips)
A = 1,
B = 2,
C = 3 ;
% 4 inferences, 0.000 CPU in 0.000 seconds (69% CPU, 222222 Lips)
A = 1,
B = 5,
C = 6 ;
% 6 inferences, 0.000 CPU in 0.000 seconds (70% CPU, 250000 Lips)
A = B, B = 2,
C = 5.
It's worth to compare with luker's suggestion in the comments regrading using findall/3. A possible implementation is:
ramdom_max(A, B, C) :-
findall(max(A,B,C), max(A,B,C), Clauses),
random_permutation(Clauses, Permutation),
member(max(A,B,C), Permutation).
Timed call:
?- time(ramdom_max(A, B, C)).
% 40 inferences, 0.000 CPU in 0.000 seconds (78% CPU, 930233 Lips)
A = 1,
B = 5,
C = 6 ;
% 2 inferences, 0.000 CPU in 0.000 seconds (50% CPU, 200000 Lips)
A = 1,
B = 2,
C = 3 ;
% 2 inferences, 0.000 CPU in 0.000 seconds (45% CPU, 250000 Lips)
A = B, B = 2,
C = 5 ;
% 4 inferences, 0.000 CPU in 0.000 seconds (62% CPU, 250000 Lips)
A = 3,
B = 4,
C = 5.
Performance is about the same in this very limited testing. But it's also a simpler and more portable solution. Knowing a bit more about the problem you want to solve would likely allow better solutions, however.