Can SAS/OR create an MPS text file?

[cuencomi]

I will anonymize the data. In what format do you want me to send?

[RobPratt]

Ideally, you can provide the three SAS data sets that appear in the READ DATA statements.

[cuencomi]

One file is too large to attach.

[cuencomi]

here are the 2 smaller files.

[cuencomi]

Even a zipped file (9.8MB) is still too large (limit 5Mb) to attach.

[RobPratt]

Can you zip the large one?  Or include only the needed observations with PDStemTravelDistance <= MaxStemDist.

[cuencomi]

where PDStemTravelDistance <153 ;

[RobPratt]

Three customers do not appear in the zip file:

'C2819','C2820','C2821'

[cuencomi]

They were all exclude because they exceed the MaxStemDist.

where PDStemTravelDistance <153 ;

[RobPratt]

OK, then I will modify the set CUSTOMERS to exclude them because otherwise the problem is infeasible.

[RobPratt]

Is it important that Assign is binary, or is it possible to split a customer's demand among multiple facilities?

[cuencomi]

This is an operational requirement of assigning a customer to exactly one facility.

[RobPratt]

OK, the other post that I linked to allows fractional Assign values and also has CUSTOMERS = FACILITIES, so that approach does not directly apply.

For your data, I reduced the set of customers (by 3) to avoid infeasibility, as mentioned earlier:

   CUSTOMERS = setof {<i,j> in CUSTOMERS_FACILITIES} i;

I used the sparse formulation (as in the "Sparse Modeling" documentation example) with only the Capacitated and SingleSource constraints:

   var Assign {CUSTOMERS_FACILITIES} binary;
   var Build  {FACILITIES}           binary;

/* declare implicit variables */ 
   impvar Total_PDStemTravelCost = sum {<i,j> in CUSTOMERS_FACILITIES} PDStemTravelCost[i,j] * Assign[i,j];
   impvar Total_FacOperCost      = sum {j in FACILITIES} FacOperCost&cap.[j] * Build[j];  
 
/******************************************************
Declare constraints on decision variables with CON
*******************************************************/ 
  
/* define constraint: sum of customer demands assigned to a facility must be less than the facility capacity */
   con Capacitated {j in FACILITIES}:
      sum {<i,(j)> in CUSTOMERS_FACILITIES} CustZipDemand[i] * Assign[i,j] <= FacCapacity&cap.[j] * Build[j];
   
/* define constraint: each customer is assigned to exactly one facility, all customers are assigned */
   con SingleSource {i in CUSTOMERS}:
      sum {<(i),j> in CUSTOMERS_FACILITIES} Assign[i,j] = 1;

With the following SOLVE statement, I got a 1% solution after 4.5 hours with SAS/OR 15.1 on my Windows laptop:

   solve with milp / relobjgap=0.01 heuristics=aggressive;

Here are the last few lines of the iteration log:

        554817   367415    205        1087826        1076877    1.02%   15785

        554893   367453    205        1087826        1076877    1.02%   15790

        557157   345573    206        1086185        1076877    0.86%   15793

NOTE: Optimal within relative gap.

NOTE: Objective = 1086185.2119.

[RobPratt]

Correction: that log was from a slower desktop.  It took 3.5 hours to get to 1% gap on my laptop:

560186 405268 130 1088541 1077326 1.04% 12540
560498 405580 130 1088541 1077326 1.04% 12545
560792 405874 130 1088541 1077326 1.04% 12550
560880 401671 131 1087946 1077326 0.99% 12551
NOTE: Optimal within relative gap.
NOTE: Objective = 1087946.3943.

[cuencomi]

Glad to see your solution worked! I will try it and hopefully I will not get an "our of memory" as before. One thing, I am using SAS/OR 14.3 which is probably not as efficient as 15.1. Thanks again.