Multisource nlp

[Santha]

Rob. 

This NLP from the old thread worked fine. It gave me 1 solution with latitude and longitude. Here is the code:

set DIMS=1..2;
set <str> CUSTOMERS;
read data STDOPT.COGModelingData into CUSTOMERS=[Sitename];

num demand {CUSTOMERS};
read data STDOPT.COGModelingData into [Sitename] Demand=SiteParameter1;

NUM Latitude {CUSTOMERS};
read data STDOPT.COGModelingData into [Sitename] Latitude=SiteLatitude;

NUM Longitude {CUSTOMERS};
read data STDOPT.COGModelingData into [Sitename] Longitude=SiteLongitude;

var X {DIMS};

min Z = sum {i in CUSTOMERS} demand[i]*GEODIST(Latitude[i],Longitude[i],X[1],X[2]);
solve;
print X;

I want to expand the logic so that the model gives me "p" number of possible facilities. I am thinking of the logic below. We need to minimize the weighted distance from "p" points to all customers.So, Here is what I did  and I know that this is not correct. Your insight would be helpful. set p=1..3; (meaning to say how many sites I want model to find out as answer): here is my code /logic below. 

var X {DIMS,p}; -- am sure this is not the right way. 

min Z = sum {i in CUSTOMERS,p} demand[i]*GEODIST(Latitude[i],Longitude[i],X[1,p],X[2,p]);

solve; print X;

Also, should we not make sure by way of constraint to make sure the model finds exactly "p" answers.?

[RobPratt]

Here is code to solve the multisource Weber problem, where m is the number of facilities.  It is based on the code I posted in the other thread, but you can modify as needed for your data and distance function.  Because the multisource problem is nonconvex, I used the multistart option for the NLP solver and also added optional code to post-process the solution to make further local improvements.

%let m = 3;
proc optmodel;
   set DIMS = 1..2;
   set CUSTOMERS;
   set FACILITIES = 1..&m;
   num a {CUSTOMERS, DIMS};
   num demand {CUSTOMERS};
   read data cdata into CUSTOMERS=[_N_] {d in DIMS} <a[_N_,d]=col('a'||d)> demand;

   num Xlb {d in DIMS} = min {i in CUSTOMERS} a[i,d];
   num Xub {d in DIMS} = max {i in CUSTOMERS} a[i,d];
   var X {FACILITIES, d in DIMS} >= Xlb[d] <= Xub[d];
   var W {CUSTOMERS, FACILITIES} >= 0;
   impvar Distance {i in CUSTOMERS, j in FACILITIES} = 
      sqrt(sum {d in DIMS}(a[i,d]-X[j,d])^2);
   min Z = sum {i in CUSTOMERS, j in FACILITIES} W[i,j]*Distance[i,j];
   con DemandCon {i in CUSTOMERS}:
      sum {j in FACILITIES} W[i,j] = demand[i];

   solve with nlp / ms;
   print X;
   put _OBJ_=;

   /* post-processing: assign each customer to closest facility */
   set CUSTOMERS_j {FACILITIES} init {};
   num minDistance, argminDistance;
   for {i in CUSTOMERS} do;
      minDistance = constant('BIG');
      argminDistance = .;
      for {j in FACILITIES} do;
         if minDistance > Distance[i,j] then do;
            minDistance = Distance[i,j];
            argminDistance = j;
         end;
      end;
      for {j in FACILITIES} W[i,j] = 0;
      W[i,argminDistance] = demand[i];
      CUSTOMERS_j[argminDistance] = CUSTOMERS_j[argminDistance] union {i};
   end;
   put _OBJ_=;

   /* post-processing: solve each facility separately */
   min SingleFacilityObjective {j in FACILITIES} = 
      sum {i in CUSTOMERS_j[j]} demand[i]*Distance[i,j];
   problem SingleFacilityProblem {j in FACILITIES} include
      {d in DIMS} X[j,d]
      SingleFacilityObjective[j];
   for {j in FACILITIES} do;
      put j=;
      use problem SingleFacilityProblem[j];
      solve;
   end;
   put (sum {j in FACILITIES} SingleFacilityObjective[j])=;
   print X;

   create data Xdata from [j]=FACILITIES {d in DIMS} <col('X'||d)=X[j,d].sol>;
   create data assignments from [j i]={j in FACILITIES, i in CUSTOMERS_j[j]} W[i,j]
      x1=a[i,1] y1=a[i,2] x2=X[j,1] y2=X[j,2];
quit;

proc sgplot data=assignments;
   vector x=x2 y=y2 / xorigin=x1 yorigin=y1 group=j;
run;

[Santha]

Rob. 

Thank you once again. I will look into it and will understand completely. 

you are awesome

[Santha · Accepted solution]

Rob.

The read data statement reads the lat and lon  and demand. 

I have few columns like SiteState, SiteBusiness etc that I want to use it for my outputs. I understand that this is not needed for the proc optmodel. So the question is , is it good to create columns in the "create data" step like sitestate, SiteBusines and update it using lat, lon and demand from the original table. But there could be 2 sites with same lat, lon and demand but only thing that is different could be it could be different SiteBusiness.  What would you recommend?

[RobPratt]

You have this code:

set <str> CUSTOMERS;
read data STDOPT.COGModelingData into CUSTOMERS=[Sitename];
num demand {CUSTOMERS};
read data STDOPT.COGModelingData into [Sitename] Demand=SiteParameter1;
NUM Latitude {CUSTOMERS};
read data STDOPT.COGModelingData into [Sitename] Latitude=SiteLatitude;
NUM Longitude {CUSTOMERS};
read data STDOPT.COGModelingData into [Sitename] Longitude=SiteLongitude;

It would be simpler to use one READ DATA statement:

set <str> CUSTOMERS;
num demand {CUSTOMERS};
NUM Latitude {CUSTOMERS};
NUM Longitude {CUSTOMERS};
read data STDOPT.COGModelingData into CUSTOMERS=[Sitename]
Demand=SiteParameter1 Latitude=SiteLatitude Longitude=SiteLongitude;

If you have additional customer attributes, you can declare them as additional (string or numeric) parameters:

str SiteState {CUSTOMERS};
str SiteBusiness {CUSTOMERS};

You can then populate these parameters however you want and use them in a CREATE DATA statement.  Alternatively, you can populate the value of a column by evaluating an expression in the CREATE DATA statement itself, without explicitly declaring new parameters, as demonstrated in the documentation.

[Santha]

Rob

THANK YOU. it worked. 

[RobPratt]

Glad to help, but it looks like you accepted your second question as the solution to your first question.