Re: GMPL/GLPK display objective function value

Hello again !

Doing more tests with this now I've got confused again, I commented out 
the "shift" term in the objective function and was expecting to get the 
same objective function but it's giving a conflicting result, doesn't it 
should have the same line in the "Objective" solution in both cases ?

====

Objective:  Reduced_Cost = -0.4666666667 (MINimum)

Objective:  Reduced_Cost = -1.466666667 (MINimum)

====

AMPL

====

# ----------------------------------------
# KNAPSACK SUBPROBLEM FOR CUTTING STOCK
# ----------------------------------------
param roll_width > 0;         # width of raw rolls

set WIDTHS;     # set of widths to be cut

param price {WIDTHS} default 0.0;

var Use {WIDTHS} /*integer*/ >= 0;

minimize Reduced_Cost:
    /*1*/ - sum {i in WIDTHS} price[i] * Use[i];

subject to Width_Limit:
    sum {i in WIDTHS} i * Use[i] <= roll_width;

data;
param roll_width := 110 ;

set WIDTHS :=
      20
      45
      50
      55
      75;
param price :=
     [20] 0.2
     [45] 0.5
     [50] 0.5
     [55] 0.5
     [75] 1
;

solve;

display Reduced_Cost, Reduced_Cost.val;

====

AMPL output:

====

myampl-ng knapsack-3.ampl
MINOS 5.51: optimal solution found.
1 iterations, objective -1.466666667
Reduced_Cost = -1.46667
Reduced_Cost = -1.46667

====

GLPSOL

====

# ----------------------------------------
# KNAPSACK SUBPROBLEM FOR CUTTING STOCK
# ----------------------------------------
param roll_width > 0;         # width of raw rolls

set WIDTHS;     # set of widths to be cut

param price {WIDTHS} default 0.0;

var Use {WIDTHS} /*integer*/ >= 0;

minimize Reduced_Cost:
    /*1*/ -sum {i in WIDTHS} price[i] * Use[i];

subject to Width_Limit:
    sum {i in WIDTHS} i * Use[i] <= roll_width;

solve;

display Reduced_Cost;

data;
param roll_width := 110 ;

set WIDTHS :=
      20
      45
      50
      55
      75;
param price :=
     [20] 0.2
     [45] 0.5
     [50] 0.5
     [55] 0.5
     [75] 1
;

====

GLPSOL output:

====

myglpsol -m knapsack-3.glpsol
GLPSOL: GLPK LP/MIP Solver, v4.65
Parameter(s) specified in the command line:
  -m knapsack-3.glpsol
Reading model section from knapsack-3.glpsol...
Reading data section from knapsack-3.glpsol...
knapsack-3.glpsol:38: warning: unexpected end of file; missing end 
statement inserted
38 lines were read
Generating Reduced_Cost...
Generating Width_Limit...
Model has been successfully generated
GLPK Simplex Optimizer, v4.65
2 rows, 5 columns, 10 non-zeros
Preprocessing...
1 row, 5 columns, 5 non-zeros
Scaling...
  A: min|aij| =  2.000e+01  max|aij| =  7.500e+01  ratio = 3.750e+00
GM: min|aij| =  1.000e+00  max|aij| =  1.000e+00  ratio = 1.000e+00
EQ: min|aij| =  1.000e+00  max|aij| =  1.000e+00  ratio = 1.000e+00
Constructing initial basis...
Size of triangular part is 1
*     0: obj =   0.000000000e+00 inf =   0.000e+00 (5)
*     1: obj =  -1.466666667e+00 inf =   0.000e+00 (0)
OPTIMAL LP SOLUTION FOUND
Time used:   0.0 secs
Memory used: 0.1 Mb (110318 bytes)
Display statement at line 20
Reduced_Cost.val = -1.46666666666667
Model has been successfully processed

====

GLPSOL solution:

====

Problem:    knapsack
Rows:       2
Columns:    5
Non-zeros:  10
Status:     OPTIMAL
Objective:  Reduced_Cost = -1.466666667 (MINimum)

    No.   Row name   St   Activity     Lower bound   Upper bound Marginal
------ ------------ -- ------------- ------------- ------------- 
-------------
      1 Reduced_Cost B       -1.46667
      2 Width_Limit  NU           110 110    -0.0133333

    No. Column name  St   Activity     Lower bound   Upper bound Marginal
------ ------------ -- ------------- ------------- ------------- 
-------------
      1 Use[20]      NL             0 0                   0.0666667
      2 Use[45]      NL             0 0                         0.1
      3 Use[50]      NL             0 0                    0.166667
      4 Use[55]      NL             0 0                    0.233333
      5 Use[75]      B        1.46667             0

Karush-Kuhn-Tucker optimality conditions:

KKT.PE: max.abs.err = 1.42e-14 on row 2
         max.rel.err = 6.43e-17 on row 2
         High quality

KKT.PB: max.abs.err = 0.00e+00 on row 0
         max.rel.err = 0.00e+00 on row 0
         High quality

KKT.DE: max.abs.err = 0.00e+00 on column 0
         max.rel.err = 0.00e+00 on column 0
         High quality

KKT.DB: max.abs.err = 0.00e+00 on row 0
         max.rel.err = 0.00e+00 on row 0
         High quality

End of output

====

Cheers !

On 29/8/20 11:32, Domingo Alvarez Duarte wrote:
> Hello Andrew !
>
> Again thank you very much for your help !
>
> After replying to your comments I reviewed again the wikibook link 
> (https://en.wikibooks.org/wiki/GLPK/Troubleshooting#Objective_shift_term_ignored_when_exported) 
> and realized that somehow I've got lost because of the of the 
> misleading AMPL result and both solvers report of the objective value.
>
> Also I think that like in the exported LP file we get a comment about 
> the constant term, the solution file also somehow should have a 
> comment about it.
>
> So in this case AMPL has a bug.
>
> I think that this function (and similares) should have a return value 
> that could tell about this:
>
> ====
>
> void glp_mpl_build_prob(glp_tran *tran, glp_prob *prob)
>
> to
>
> int glp_mpl_build_prob(glp_tran *tran, glp_prob *prob)
>
> {
>
> int ret = 0;
>
> ...
>
> if (mpl_get_row_c0(tran, i) != 0.0) {
>             xprintf("glp_mpl_build_prob: row %s; constant term %.12g ig"
>                "nored\n",
>                mpl_get_row_name(tran, i), mpl_get_row_c0(tran, i));
>
>             ret = WARNING_CONST_TERM_IGNORED;
>
> }
>
> ...
>
> return ret;
>
> }
>
> ====
>
> Cheers !
>
> On 29/8/20 11:01, Domingo Alvarez Duarte wrote:
>> Hello Andrew !
>>
>> Thanks for reply !
>>
>> The problem is that the GMPL is not informed/knows anything about the 
>> warning/divergence in the results and goes blindly.
>>
>> Te exit code of glpsol also doesn't reflect anything that could be 
>> detected, only by visual inspection of the output or redirect the 
>> output and scan it for possible warnings.
>>
>> Somehow GMPL should have a way to acknowledge situations like these.
>>
>> ===
>>
>> >Exit code: 0
>>
>> ===
>>
>> Cheers !
>>
>> On 29/8/20 0:46, Andrew Makhorin wrote:
>>>> While trying to implement multi solve statements I found that in GMPL
>>>> the display of an objective function after solving do not show the
>>>> optimal value.
>>> BTW, glpsol warns about that:
>>>
>>> [...]
>>> 38 lines were read
>>> Generating Reduced_Cost...
>>> Generating Width_Limit...
>>> Model has been successfully generated
>>> glp_mpl_build_prob: row Reduced_Cost; constant term 1 ignored
>>> [...]