Re: [eclipse-clp-users] Instantiation fault in bb_min with non-default strategy

On 13/03/15 19:39, Sergii Dymchenko wrote:
> Hi,
> I wrote a blog post about solving PuzzlOR "Electrifying" puzzle with ECLiPSe:
> http://sdymchenko.com/blog/2015/03/13/electrifying-eclipse/
>
> While solving the puzzle I encountered and issue with bb_min and strategies
> other than the default one.
>
> Consider this program (basically the same program as in the blog post linked):
>
> :- lib(ic).
> :- lib(branch_and_bound).
> solve(HouseXs, HouseYs, K, GenXs, GenYs, Cost) :-
>      dim(HouseXs, [N]),
>      MaxX #= max(HouseXs), MaxY #= max(HouseYs),
>      dim(GenXs, [K]), dim(GenYs, [K]),
>      GenXs :: 1..MaxX, GenYs :: 1..MaxY,
>      ( for(I, 1, N), foreach(Di, Distances), param(HouseXs, HouseYs, GenXs, > GenYs, K) do
>          ( for(J, 1, K), fromto(1.0Inf, Dprev, Dcurr, Di), param(I, HouseXs, > HouseYs, GenXs, GenYs) do
>              Dcurr $= min(Dprev, sqrt(sqr(HouseXs[I] - GenXs[J]) + > sqr(HouseYs[I] - GenYs[J])))
>          )
>      ),
>      Cost $= sum(Distances),
>      bb_min(search([](GenXs, GenYs), 0, most_constrained, indomain_middle, > complete, []), Cost, bb_options{delta:0.1, strategy:dichotomic}).
>
> main :-
>      HouseXs = [](2, 2, 3, 3, 4, 4,  5, 5, 5, 5, 6, 6, 7, 7, 8, 8, 8, 9, 9, 10),
>      HouseYs = [](2, 5, 6, 8, 2, 10, 2, 3, 6, 8, 1, 5, 2, 8, 5, 7, 8, 4, 7, 3),
>      solve(HouseXs, HouseYs, 3, GenXs, GenYs, _Cost),
>      ( foreacharg(X, GenXs), foreacharg(Y, GenYs) do
>          LetterCode is 0'A + Y - 1,
>          char_code(Letter, LetterCode),
>          write(X), write(Letter), nl
>      ).
>
> With strategy:dichotomicI get:
>
> Found a solution with cost 67.5083636407248__67.50836364072498
> Found no solution with cost 0.0 .. 33.75418182036249
> instantiation fault in 33.75418182036249 =< _325826{49.307233137943534 ..
> 49.30723313794369}
> abort
>
> There is a similar issue with strategy:restart, but the program works OK with
> the default continuestrategy.
> Is this an ECLiPSe bug?


Not strictly a bug, but an unhelpful error message.  bb_min expects the search 
routine to instantiate the Cost variable, which, as often with continuous 
variables, isn't the case here.

What you have to do is instantiate the Cost variable to a bounded-real constant, 
after the search has finished, like so:

      bb_min((
               search(...),
               get_var_bounds(Cost,CostL,CostH),
               breal_from_bounds(CostL,CostH,Cost)
           ),
           Cost,
           bb_options{delta:0.1, strategy:dichotomic} )



-- Joachim