I´m trying to build a predicate pattern(List,Pattern) that takes a List formed only by a repeated pattern and the output has to be that pattern.
Some examples of the List:
List1=[a,b,a,b]
List2=[1,2,3,1,2,3]
List3=[a,a,a,a,a,a,a,a]
As you can see, in each case either the list and the pattern can have different lenghts.
And the output in each case would be:
Pattern1=[a,b]
Pattern2=[1,2,3]
Pattern3=[a]
The only way I can think about a solution is taking the first element of the List (for example, in List2 would be "1") and going through List2 until I find again a "1" and then put in Pattern everything before the second 1 ("123"), but I don´t think it is the best solution. Does anybody know an easier way to solve it? Maybe with Append/3 or Member/2? Thank you!
You are looking for the shortest sequence Q ("pattern") such that list L is n > 0 concatenations of Q (whereby if n = 1 iff Q = L), then
If you have a verifying predicate which verifies that L is indeed a concatenation of a (non necessarily) shortest Q:
multiple_concatenations(X,X). % L = 1 * Q
multiple_concatenations(Q,L) :- % L = N * Q (for some N >= 1, Q <> []) if
concatenation(Q,Rest,L), % L = Q + Rest and
multiple_concatenations(Q,Rest). % Rest = M * Q (for some M)
Where concatenation/3 is just the sanely named append/3 from Prolog:
concatenation(List1,List2,List12) :-
append(List1,List2,List12).
... then you can try to find a shortest Q by just generating longer and longer potential _Q_s (of length going from 1 to length(L)) and break off at the first Q which passes multiple_concatenations(Q,L,N):
shortest_q(Q,L) :-
length(L,Length), % The length of L
must_be(positive_integer,Length), % Enforce length(L) > 0; throws if not
length(Q,N), % Generate a longer and longer
% "list template" Q (i.e. a list with only
% uninstantiated variables) on backtracking,
% with N = 0,1,2,3,...
N>0, % Don't want an empty template though
(concatenation(Q,_,L) % Q's uninstantiated members are
% now instantiated; if this fails,
->
(multiple_concatenations(Q,L), % Check whether Q is acceptable
!) % If yes, cut (i.e. break off at first answer)
;
fail). % If concatenation(Q,_,L) fails, fail the
% predicate: we have
% gone too far (Q is longer than L)
Add a few plunit test cases, which are doubleplus important in the "what am I computing right now?" Prolog wonderland:
:- begin_tests(mq).
test(1) :-
shortest_q(Q,[a,b,a,b]),
assertion(Q == [a,b]).
test(2) :-
shortest_q(Q,[1,2,3,1,2,3]),
assertion(Q == [1,2,3]).
test(3) :-
shortest_q(Q,[a,a,a,a,a,a,a,a]),
assertion(Q == [a]).
test(4) :-
shortest_q(Q,[a,b,c,d,e,f,g]),
assertion(Q == [a,b,c,d,e,f,g]).
:- end_tests(mq).
And so:
?- run_tests.
% PL-Unit: mq .... done
% All 4 tests passed
true.
Note however that "verification mode" accepts a sequence longer than the minimum:
?- shortest_q([1,2,3],[1,2,3,1,2,3]).
true.
?- shortest_q([1,2,3,1,2,3],[1,2,3,1,2,3]).
true.
A simple solution using only append/3 is:
% pattern(+List, -Pattern)
pattern([], _). % Any pattern repeated 0 times gives []
pattern(L, [P|Ps]) :- % [P|Ps] guarantees a non-empty pattern
append([P|Ps], R, L), % Gets a prefix of L as a possible pattern
pattern(R, [P|Ps]), % Checks whether prefix is indeed a pattern
!. % stops when the shortest pattern is found
Examples:
?- pattern([a,b,a,b], P).
P = [a, b].
?- pattern([1,2,3,1,2,3], P).
P = [1, 2, 3].
?- pattern([a,a,a,a,a,a,a], P).
P = [a].
Related
I want to develop a predicate in prolog called next, which given a list returns another list with the N elements following the last value of the list where N is the size of the main list. For example: next([1,2,3,4], R).
will return R = [5,6,7,8]. or: next([11,12,13], R). It will return R = [14,15,16].
The problem I have is that if I iterate over the main list until I am left with its last element and start adding the next one to it to the result list, I don't know how many times I should iterate since I don't know what the length of the main list was.This is why my algorithm goes into a loop.
next([], []).
next([X], [X1|Res]) :- X1 is X + 1, next3([X1],Res),!.
next([H|T], [X]) :- next3(T, X).
How about this.
Here we use Constraint Logic Programming to constraint the elements of the result list to be increasing-monotonically-by-1 but just set them to actual values (in one instruction) once we know the last element of the input list.
We also use an open list (a list with an unbound fin) instead of append/3 to grow the output list at its end efficiently. This idiom called "using a difference list".
Finally we add test cases.
:- use_module(library(clpfd)).
% ---
% Called by the customer
% ---
nextor([],[]) :- !. % Empty input list means no work to do!
nextor([X|Xs],Out) :- % The input list contains at least 1 element
assertion(integer(X)), % This is for stating what we assume
OpenList = [K|Fin], % We will create a list of fresh variables; in order
% _to append easily, the list is kept "open", i.e. its tail end
% _is not [] as in a proper list but an unbound variable
% _(namely, Fin). The first element is as yet undefined,
% it is the fresh variable K.
assertion(\+is_list(OpenList)), % Not a list at present time.
nextor_w(X,Xs,[K|Fin],LastFin,LastX), % Processing the list with the first element X already
% _separated out (for assertions). To grow the OpenList at
% _its end, we just need Fin (we don't care about that list's
% Tip when we grow it at the end). Finally, to communicate
% the last Fin set up in the depth of the recursion to this
% call place, use LastFin. The last X found will be in LastX.
LastFin=[], % The open list is close (made proper list) by setting its
% _final Fin to [].
assertion(is_list(OpenList)), % Yes, it is a list now!
K #= LastX+1, % Now that LastX is known, we know K too.
% _The constraint propagates down the list, fixing the still
% unbound variables in OpenList (which is now a closed list).
Out = OpenList. % Unify for result.
% ---
% Does the recursion down the input list
% ---
nextor_w(Xp,[],[_|Fin],Fin,Xp) :- !. % At the end of recursion, communicate the "last X" and the
% and the "latest Fin" back to the caller.
nextor_w(Xp,[X|Xs],[K|Fin],FinOut,XpOut) :-
assertion(Xp+1 =:= X), % The input list is assumed to increase monotonously.
Kn #= K+1, % Next K is constrained to be previous K + 1,
Fin = [Kn|NewFin], % The Fin of the open list is set to a new list cell, with a new Fin
nextor_w(X,Xs,[Kn|NewFin],FinOut,XpOut).
% ---
% Testing
% ---
:- begin_tests(nextor).
test("empty list" ,true(R == [])) :-
nextor([],R).
test("1-elem list",true(R == [2])) :-
nextor([1],R).
test("2-elem list",true(R == [3,4])) :-
nextor([1,2],R).
test("3-elem list",true(R == [4,5,6])) :-
nextor([1,2,3],R).
:- end_tests(nextor).
And so:
?- run_tests.
% PL-Unit: nextor .... done
% All 4 tests passed
true.
How about doing:
next(In,Out) :-
length(In,N),
maplist(plus(N),In,Out).
Uses length/2 and maplist/3. This works for the examples in your question - next([1,2,3,4],R). and next([11,12,13],R). - but only because the lists contain consecutive numbers. next([23,2,18],R). will unify R with [26,5,21].
Or:
next(In,Out) :-
length(In,N),
last(In,LastValue),
MinValue is LastValue+1,
MaxValue is LastValue+N,
numlist(MinValue,MaxValue,Out).
Uses last/2 and numlist/3. With this approach next([23,2,18],R). will unify R with [19,20,21].
This works fine for me:
next(Xs,Ys) :-
next(Xs,[_],Ys).
next([_],[],[]).
next([X],[_|Q],[X1|R]) :-
X1 is X + 1,
next([X1],Q,R).
next([_|T],Q,R) :-
next(T,[_|Q],R).
As it runs down the list it is building up a second list. When it finds the end of the first list it then runs down the second list building the output.
I got this kind of output:
?- next([1,2,3,7,8],W).
W = [9, 10, 11, 12, 13] .
I have a list of variables E and a list L and I want a predicate that works like this:
E=[A,B,C,D]
L=[(1,b),(3,m)]
solve(E,L).
E=[b,B,m,D]
Basically solve() should run through the list L and change E by using (a,b) to unify the variable at index a with the atom B. Is there any way to do this?
The meaning of the (badly named) solve/2 predicate is something like "for every pair (Index, Element), the Index-th element of the input list is Element". You are likely using a Prolog implementation that already has a predicate called something like nth1/3 which expresses "the Index-th element of List is Element". For example, in SWI-Prolog:
?- List = [A, B, C, D], nth1(3, List, this_is_the_third_element).
List = [A, B, this_is_the_third_element, D],
C = this_is_the_third_element.
So an alternative implementation of your predicate simply calls nth1/3 for each of your (Index, Element) pairs:
solve(_List, []).
solve(List, [(Index, Elem) | Pairs]) :-
nth1(Index, List, Elem),
solve(List, Pairs).
And with this you're done:
?- E = [A, B, C, D], L = [(1, b), (3, m)], solve(E, L).
E = [b, B, m, D],
A = b,
C = m,
L = [(1, b), (3, m)] ;
false.
Note that this solution is simple, but it has quadratic complexity in the length of the input list: nth1/3 might have to visit the entire N-element list N times. In the unlikely case that you need this predicate for a performance-critical part of some larger program, consider the more optimized solution sketched in the other answer.
Is there any way to do this?
Certainly. And as they say in Perl: "There is more than one way to do it".
Couple of problems:
Do not use (1,b). Use the idiomatic -(1,b) instead, which is written as 1-b (the pair). This gives you a list of pairs: L=[1-b,3-m]. There is a library specifically dealing with such pairs: https://www.swi-prolog.org/pldoc/man?section=pairs - alternatively you can use real maps implemented with AVL trees: https://www.swi-prolog.org/pldoc/man?section=assoc
Now you just need to:
sort the list of pairs, probably using keysort: https://www.swi-prolog.org/pldoc/doc_for?object=sort/2 or https://www.swi-prolog.org/pldoc/doc_for?object=sort/4
Go through the list left to right, keeping the current index, and performing a replacement when the next key in your sorted list is hit, or just retaining the existing term from the list otherwise. The result goes into an accumulator variable as head of a list.
Done! Special handling of out-of-bounds indexes etc. to be suitably handled by throwing or failing.
How to go through the sorted list of pairs (I didn not test this!):
% case of Index hit:
go_through([Index-Value|Rest],Index,InList,OutList) :-
InList = [I|Rest],
OutList = [Value|More],
succ(Index,NextIndex),
go_through(Rest,NextIndex,Rest,More).
% case of Index miss:
go_through([NotYetIndex-Value|Rest],Index,InList,OutList) :-
NotYetIndex > Index, % that should be the case
InList = [I|Rest],
OutList = [I|More],
succ(Index,NextIndex),
go_through(Rest,NextIndex,Rest,More).
go_through([],_,L,L). % DONE
Alternatively, you can write a replace0 that replaces-by-index in a list, and go through the L list.
Addendum: Working code using go_through
Actually contains a few subtlties
another_vectorial_replace1(ListIn,ReplacePairs,ListOut) :-
maplist([_,_]>>true,ListIn,ListOut), % Bonus code: This "makes sure" (i.e. fails if not)
% that ListIn and ListOut are the same length
maplist([(A,B),A-B]>>true,ReplacePairs,RealPairs), % Transform all (1,b) into [1,b]
maplist([K-_]>>integer(K),RealPairs), % Make sure the RealPairs all have integers on first place
keysort(RealPairs,RealPairsSorted), % Sorting by key, which are integers; dups are not removed!
debug(topic,"ListIn: ~q",[ListIn]),
debug(topic,"RealPairsSorted: ~q",[RealPairsSorted]),
go_through(RealPairsSorted,1,ListIn,ListOut),
debug(topic,"ListOut: ~q",[ListOut]).
% Case of Index hit, CurIndex is found in the first "Replacement Pair"
go_through([CurIndex-Value|RestPairs],CurIndex,ListIn,ListOut) :-
!, % Commit to choice
ListIn = [_|Rest],
ListOut = [Value|More],
succ(CurIndex,NextIndex),
go_through(RestPairs,NextIndex,Rest,More).
% Case of Index miss:
go_through([NotYetIndex-V|RestPairs],CurIndex,ListIn,ListOut) :-
NotYetIndex > CurIndex, % that should be the case because of sorting; fail if not
!, % Commit to choice
ListIn = [X|Rest],
ListOut = [X|More],
succ(CurIndex,NextIndex),
go_through([NotYetIndex-V|RestPairs],NextIndex,Rest,More).
% Case of DONE with list traversal
% Only succeed if there are not more pairs left (i.e. no out-of-bound replacements)
go_through([],_CurIndex,L,L).
% ===
% Tests
% ===
:- begin_tests(another_vectorial_replace1).
test(empty) :- another_vectorial_replace1([],[],LO),
LO=[].
test(nop_op) :- another_vectorial_replace1([a,b,c,d],[],LO),
LO=[a,b,c,d].
test(one) :- another_vectorial_replace1([a],[(1,xxx)],LO),
LO=[xxx].
test(two) :- another_vectorial_replace1([a,b,c,d],[(4,y),(2,x)],LO),
LO=[a,x,c,y].
test(full) :- another_vectorial_replace1([a,b,c,d],[(1,e),(2,f),(3,g),(4,h)],LO),
LO=[e,f,g,h].
test(duplicate_replacement,[fail]) :- another_vectorial_replace1([a],[(1,x),(1,y)],_).
test(out_of_bounds_high,[fail]) :- another_vectorial_replace1([a],[(2,y)],_).
test(out_of_bounds_low,[fail]) :- another_vectorial_replace1([a],[(0,y)],_).
:- end_tests(another_vectorial_replace1).
rt :- debug(topic),run_tests(another_vectorial_replace1).
Addendum 2
Replacement using maplist/N, foldl/N and library(assoc)
Recursive calls disappear behind the curtain!
https://github.com/dtonhofer/prolog_notes/blob/master/code/vector_replace0.pl
(the following assumes that the indices in the pairs list will be sorted, in increasing order, as the example in the question indicates.)
What you said can be written as one conjunction
E=[A,B,C,D], L=[(1,a),(3,c)], solve(E,L), E=[a,B,c,D].
which you intend to be holding under the proper definition of solve/2 that you seek to find. But isn't it like saying
E=[A|E2], L=[(1,a)|L2],
E2=[B,C,D], L2=[(3,c)],
solve(E2,L2), E2=[B,c,D],
E=[a|E2].
? Although, something doesn't quite fit right, here. c in E2 appears in second position, not 3rd as indicated by its entry in L2.
But naturally, L2 must be indexed from 2, since it is a tail of L which is indexed from 1. So we must make this explicit:
E=[A,B,C,D], L=[(1,a),(3,c)], solve(E,L), E=[a,B,c,D]
==
E=[A,B,C,D], L=[(1,a),(3,c)], solve(E,1,L), E=[a,B,c,D] % starting index 1
==
E=[A|E2], L=[(1,a)|L2],
E2=[B,C,D], L2=[(3,c)],
solve(E2,2,L2), E2=[B,c,D], E=[a|E2]
must, and now can, hold. But where did a get from, in E? What we actually mean here is
E=[A|E2], L=[(1,a)|L2],
p( (1,a), 1, a), % index match
E2=[B,C,D], L2=[(3,c)],
solve(E2,2,L2), E2=[B,c,D], % starting index 2
E=[a|E2]
with p/3 defined as
p( (I,A), I, A).
And so it must also hold that
E2=[B|E3], L2=[(3,c)],
\+ p( (3,c), 2, c), % index mismatch
E3=[C,D], L3=L2,
solve(E3,3,L3), E3=[c,D], E2=[B|E3]
L2 is not traversed along at this step (L3=L2), since p( (3,c), 2, c) does not hold.
Do you see how the recursive definition of solve/3 reveals itself here? Could you finish it up?
I'm trying to write a palindrome function in Prolog. I know I could just use something like
palindrome(List) :- reverse(List, List).
But I'm trying to figure out a way without using the built in reverse. I've created my own reverse rule:
rev([], []).
rev([H|T], X) :- rev(T, Y), append(Y, [H], X).
And what I'd like is, given a list, say [a,b,c,d], I'd like to do something like "X = rev([a,b,c,d]), but I'm really not sure whether this is possible in Prolog.
If it is, the way I would write my palindrome function would be something like:
palindrome(List) :- append(L1, rev(L1), List).
Is it possible to do what I'm trying to do - i.e. X = rev([a,b,c,d])?.
Thanks.
Palindromes are lists that read the same from front to back and from back to front. So the example you have given, [a,b,c,d] and it's reversal, constitute a palindrome if the first is directly followed by the second: [a,b,c,d,d,c,b,a]. Since you are trying to describe specific kinds of lists, it is very tempting to use Prolog DCGs for the task. With them you can define palindromes like so:
palindrome(X) :-
phrase(palindrome,X).
palindrome --> % base case for even number of elements
[].
palindrome --> % base case for odd number of elements
[A].
palindrome --> % general case: a palindrome is
[A], % some element A...
palindrome, % ... followed by a palindrome ...
[A]. % ... followed by element A
The most general query is producing palindromes with variables for each position:
?- palindrome(P).
P = [] ? ;
P = [_A] ? ;
P = [_A,_A] ? ;
P = [_A,_B,_A] ? ;
P = [_A,_B,_B,_A] ? ;
P = [_A,_B,_C,_B,_A] ?
...
Or alternatively you can test if a specific list is a palindrome:
?- palindrome("rats live on no evil star").
yes
?- palindrome([1,2,3,2,1]).
yes
?- palindrome([a,b,c,d]).
no
?- palindrome([a,b,c,d,d,c,b,a]).
yes
If you insist on using list reversal you can define the relation like so:
list([]) -->
[].
list([X|Xs]) -->
[X],
list(Xs).
invlist([]) -->
[].
invlist([X|Xs]) -->
invlist(Xs),
[X].
palindrome --> % a paindrome is
list(L), % a list followed
invlist(L). % by its reversal
palindrome --> % a palindrome is
list(L), % a list followed by
[_A], % some element
invlist(L). % then by the reversed list
The first of the above queries produces the answers in a different order now, namely the solutions with an even number of elements first:
?- palindrome(P).
P = [] ? ;
P = [_A,_A] ? ;
P = [_A,_B,_B,_A] ? ;
P = [_A,_B,_C,_C,_B,_A] ?
...
The other example queries yield the same result. However, the first definition seems to be clearly preferable to me. Not only because it is shorter as there is no need for additional DCG rules but also because it is producing the results in a fair order: empty list, one element, two elements, ... With the second version you get all the lists with an even number of elements first and there are infinitely many of those. So you never get to see a solution with an odd number of elements with the most general query.
I am trying to make use of prolog predicates and find middle element of a given list. My idea was to cut first and last element of list using recursion.Unfortunately I dont know how to handle recursion call properly.
delete_last(L, L1) :-
append(L1, [_], L).
delete_first(L,L1) :-
append([_],L1,L).
check_len(L) :-
length(L,LEN), \+ 1 is LEN.
delete_both([],_):-
false.
delete_both([_,_],_) :-
false.
delete_both([X],X):-
true, write('MidElement').
delete_both(L,L2) :-
delete_first(LT,L2), delete_last(L,LT),check_len(LT)
->write('here should be recursive call only when length is more than one').
I would be grateful for any help.
It would save a lot of typing if you checked the length of the list, calculated the position of the middle element, and only then traversed the list to get the element at that position. With SWI-Prolog, this would be:
?- length(List, Len),
divmod(Len, 2, N, 1),
nth0(N, List, a).
List = [a], Len = 1, N = 0 ;
List = [_G2371, a, _G2377], Len = 3, N = 1 ;
List = [_G2371, _G2374, a, _G2380, _G2383], Len = 5, N = 2 . % and so on
This solution makes sure the list has an odd length. You can see the documentation of divmod/4 if you need to define it yourself. Or, if the list does not have to have and odd, length, just use N is Len div 2. If for some reason you are not allowed to use nth0/3, it is still an easier predicate to implement than what you are trying to do.
You can tighten up what you have quite a bit as follows:
delete_last(L, L1) :-
append(L1, [_], L).
delete_first([_|L], L).
% No need to check length of 1, since we only need to check
% if L = [X] in the caller, so we'll eliminate this predicate
%check_len(L) :-
% length(L, 1). % No need for an extra variable to check length is 1
% Clauses that yield false are not needed since clauses already fail if not true
% So you can just remove those
%
delete_both([X], X) :-
write('MidElement').
% Here you need to fix the logic in your main clause
% You are deleting the first element of the list, then the last element
% from that result and checking if the length is 1.
delete_both(L, X) :-
delete_first(L, L1), % Remove first and last elements from L
delete_last(L1, LT),
( LT = [X] % Check for length of 1
-> true
; delete_both(LT, X) % otherwise, X is result of delete_both(LT, X)
).
With results:
| ?- delete_both([a,b,c,d,e], X).
X = c
yes
| ?- delete_both([a,b,c,d,e,f], X).
no
A DCG solution also works well here:
% X is the middle if it is flanked by two sequences of the same length
%
middle(X) --> seq(N), [X], seq(N).
seq(0) --> [].
seq(N) --> [_], { N #= N1 + 1 }, seq(N1).
middle(List, X) :- phrase(middle(X), List).
With results:
| ?- middle([a,b,c,d,e], X).
X = c ? ;
(1 ms) no
| ?- middle(L, a).
L = [a] ? ;
L = [_,a,_] ? ;
L = [_,_,a,_,_] ?
...
Another possible solution is to use SWI Prolog's append/2 predicate, which appends a list of lists (assuming you're using SWI):
middle(L, X) :-
same_length(Left, Right),
append([Left, [X], Right], L).
same_length([], []).
same_length([_|T1], [_|T2]) :- same_length(T1, T2).
In all of the above solutions, the predicate fails if the list has an even number of elements. Since that's what your original solution does, I assumed that's what is required. If there is a specific requirement for even lists, that needs to be stated clearly.
As a Prolog newbie, I try to define a predicate filter_min/2 which takes two lists to determine if the second list is the same as the first, but with all occurrences of the minimum number removed.
Sample queries with expected results:
?- filter_min([3,2,7,8], N).
N = [3,7,8].
?- filter_min([3,2,7,8], [3,7,8]).
true.
I tried but I always get the same result: false. I don't know what the problem is. I need help!
Here is my code:
filter_min(X,Y) :-
X == [],
write("ERROR: List parameter is empty!"),
!;
min_list(X,Z),
filter(X,Y,Z).
filter([],[],0).
filter([H1|T1],[H2|T2],Z) :-
\+ number(H1),
write("ERROR: List parameter contains a non-number element"),
!;
H1 \= Z -> H2 is H1, filter(T1,T2,Z);
filter(T1,T2,Z).
There are a couple of problems with your code:
filter([],[],0). will not unify when working with any list that does not have 0 as its minimum value, which is not what you want. You want it to unify regardless of the minimum value to end your recursion.
The way you wrote filter([H1|T1],[H2|T2],Z) and its body will make it so that the two lists always have the same number of elements, when in fact the second one should have at least one less.
A correct implementation of filter/3 would be the following:
filter([],[],_).
filter([H1|T1],L2,Z):-
\+ number(H1),
write("ERROR: List parameter contains a non-number element"),
!;
H1 \= Z -> filter(T1,T2,Z), L2 = [H1|T2];
filter(T1,L2,Z).
A bounty was offered...
... for a pure solution that terminates for (certain) cases where neither the length of the first nor of the second argument is known.
Here's a candidate implementation handling integer values, built on clpfd:
:- use_module(library(clpfd)).
filter_min(Xs,Ys) :-
filter_min_picked_gt(Xs,_,false,Ys).
filter_min_picked_gt([] ,_,true ,[]).
filter_min_picked_gt([Z|Xs],M,Picked,[Z|Zs]) :-
Z #> M,
filter_min_picked_gt(Xs,M,Picked,Zs).
filter_min_picked_gt([M|Xs],M,_,Zs) :-
filter_min_picked_gt(Xs,M,true,Zs).
Some sample queries:
?- filter_min([3,2,7,8],[3,7,8]).
true ; false. % correct, but leaves choicepoint
?- filter_min([3,2,7,8],Zs).
Zs = [3,7,8] ; false. % correct, but leaves choicepoint
Now, some queries terminate even though both list lengths are unknown:
?- filter_min([2,1|_],[1|_]).
false. % terminates
?- filter_min([1,2|_],[3,2|_]).
false. % terminates
Note that the implementation doesn't always finitely fail (terminate) in cases that are logically false:
?- filter_min([1,2|_],[2,1|_]). % does _not_ terminate
For a Prolog newbie, better start with the basics. The following works when first argument is fully instantiated, and the second is an uninstantiated variable, computing the result in one pass over the input list.
% remmin( +From, -Result).
% remmin([],[]). % no min elem to remove from empty list
remmin([A|B], R):-
remmin(B, A, [A], [], R). % remove A from B to get R, keeping [A]
% in case a smaller elem will be found
remmin([C|B], A, Rev, Rem, R):-
C > A -> remmin(B, A, [C|Rev], [C|Rem], R) ;
C==A -> remmin(B, A, [C|Rev], Rem, R) ;
C < A -> remmin(B, C, [C|Rev], Rev, R).
remmin([], _, _, Rem, R) :- reverse(Rem, R).
First, we can get the minimum number using the predicate list_minnum/2:
?- list_minnum([3,2,7,8],M).
M = 2.
We can define list_minnum/2 like this:
list_minnum([E|Es],M) :-
V is E,
list_minnum0_minnum(Es,V,M).
list_minnum0_minnum([],M,M).
list_minnum0_minnum([E|Es],M0,M) :-
M1 is min(E,M0),
list_minnum0_minnum(Es,M1,M).
For the sake of completeness, here's the super-similar list_maxnum/2:
list_maxnum([E|Es],M) :-
V is E,
list_maxnum0_maxnum(Es,V,M).
list_maxnum0_maxnum([],M,M).
list_maxnum0_maxnum([E|Es],M0,M) :-
M1 is max(E,M0),
list_maxnum0_maxnum(Es,M1,M).
Next, we use meta-predicate tfilter/3 in tandem with dif/3 to exclude all occurrences of M:
?- M=2, tfilter(dif(M),[2,3,2,7,2,8,2],Xs).
Xs = [3,7,8].
Put the two steps together and define min_excluded/2:
min_excluded(Xs,Ys) :-
list_minnum(Xs,M),
tfilter(dif(M),Xs,Ys).
Let's run some queries!
?- min_excluded([3,2,7,8],Xs).
Xs = [3,7,8].
?- min_excluded([3,2,7,8,2],Xs).
Xs = [3,7,8].