About RobPratt

RobPratt · ‎02-11-2016

First, let's make sure the model is correct. I see a couple of issues. The "r in regimen" should not appear on both sides of this declaration: impvar regimen_tot_ps{r in regimen}=sum{m in month, r in regimen} ps[r,m]; The "m in month" should not appear on both sides of this declaration: impvar drug_share{d in drug, m in month} = sum{m in month, r in regimen}ps[r,m]*weight[r,d]; Is there a reason for the outer SUM here? impvar avg_dot_1=sum(sum{m in month, r in regimen} (ps[r,m]*dot[r])); After correcting these, do the results from EXPAND look like what you expect? As another check on the correctness of the model, you should probably FIX the variables to their values from the Excel solution and then call the solver.

RobPratt · ‎02-09-2016

PROC SURVEYREG is part of SAS/STAT, so you will probably get a better response by posting in the SAS Statistical Procedures community: https://communities.sas.com/t5/SAS-Statistical-Procedures/bd-p/statistical_procedures

RobPratt · ‎02-09-2016

Yes, the paper will be available online as part of the conference proceedings. Until then, here are a few other links to resources that might be helpful for you to learn PROC OPTMODEL: Papers and slides: http://support.sas.com/rnd/app/or/papers/index.html Official documentation, especially Mathematical Programming Examples: http://support.sas.com/documentation/onlinedoc/or/index.html PROC OPTMODEL course: https://support.sas.com/edu/schedules.html?id=2616&ctry=US “OPERATIONS RESEARCH WITH SAS” blog: http://blogs.sas.com/content/operations/

RobPratt · ‎02-09-2016

I'm glad to help. I should also mention that I will be presenting a paper on multiple solutions in PROC OPTMODEL at SAS Global Forum 2016 in April: http://www.sas.com/en_us/events/sas-global-forum/sas-global-forum-2016.html

RobPratt · ‎02-09-2016

You will get that error even if you just declare continuous variables. Please try commenting out those declarations and any other statements that use those variables.

RobPratt · ‎02-09-2016

Direct support for mixed integer nonlinear problems in PROC OPTMODEL is on our road map. Until then, you might consider using PROC OPTLSO: http://support.sas.com/documentation/cdl/en/orlsoug/68155/HTML/default/viewer.htm#orlsoug_hplso_toc.htm Also, you can model some nonlinearity with integer variables in PROC OPTMODEL by using the ELEMENT predicate in the CLP solver: http://support.sas.com/documentation/cdl/en/ormpug/68156/HTML/default/viewer.htm#ormpug_clpsolver_examples07.htm

RobPratt · ‎02-09-2016

The CLP solver in PROC OPTMODEL can return all or some specified number of optimal solutions. You can use either the FINDALLSOLNS option or the MAXSOLNS= option in the SOLVE WITH CLP statement: http://support.sas.com/documentation/cdl/en/ormpug/68156/HTML/default/viewer.htm#ormpug_clpsolver_syntax03.htm

RobPratt · ‎01-03-2016

That number of lines is quite unusual, and I suspect that you aren't exploiting the full capabilities of PROC OPTMODEL, which provides a very expressive modeling language that typically requires at most hundreds of lines even for complicated user-customized algorithms that call multiple solvers. It is hard to recommend anything specific without seeing more detail. Perhaps it would be best to submit your code and log to technical support, and I'll take a look.

RobPratt · ‎12-21-2015

Please post complete code and data, and I'll take a look.

RobPratt · ‎12-18-2015

For network problems, adding dummy nodes and arcs with high cost is the usual approach to diagnose infeasibility, as you have described. But since you are having trouble with that approach, you might consider using the IIS detection feature that diagnoses infeasibility for linear programming problems: http://support.sas.com/documentation/cdl/en/ormpug/68156/HTML/default/viewer.htm#ormpug_lpsolver_details15.htm This method requires an explicit LP formulation, and you can use the code in the following example, which solves a generic minimum-cost network flow problem: http://support.sas.com/documentation/cdl/en/ormpug/68156/HTML/default/viewer.htm#ormpug_lpsolver_examples05.htm You should replace the SOLVE statement with: solve with LP / iis=on; expand / iis;

RobPratt · ‎12-10-2015

Supply of 0 at a node just forces inflow to be equal to outflow at that node. There is no upper bound for the flow on each arc, but the minimization objective will drive it to be as small possible, according to the arc costs. If you post your simple example, I'll take a look. One thing you might have missed is that the two-way arcs should be replaced with two one-way arcs in the input data. The undirected version of the problem is addressed in Application 19.15 of Network Flows. The solution approach does compute all-pairs shortest paths as the first step (as you mentioned), and this is also available in PROC OPTNET or PROC OPTMODEL. Note here that the shortest paths are with respect to the sum of arc distances, not the number of arcs. But the second step is a weighted matching problem on a complete graph consisting of the odd-degree nodes, where the new cost of edge (i,j) is the shortest path distance between i and j in the original graph. You can solve this problem by using the MILP solver in PROC OPTMODEL. There is a binary decision variable for each edge, and if the solver returns a value of 1 for edge (i,j), it means that all the arcs in the shortest path between i and j in the original network should be added. In fact, you can call the network solver and then the MILP solver in the same PROC OPTMODEL session.

RobPratt · ‎12-09-2015

As described in Application 19.14 of Network Flows (Ahuja, Magnanti, Orlin), the directed version of the problem can be formulated as a minimum-cost network flow problem with 0 supply at each node and a lower bound of 1 unit of flow on each arc. You can use either PROC OPTNET... http://support.sas.com/documentation/cdl/en/ornoaug/68159/HTML/default/viewer.htm#ornoaug_optnet_details23.htm ...or the network solver in PROC OPTMODEL: http://support.sas.com/documentation/cdl/en/ormpug/68156/HTML/default/viewer.htm#ormpug_networksolver_details19.htm

RobPratt · ‎12-09-2015

Can you please share your data and code, or at least some sample data? By the way, although your problem is not TSP, note that the TSP algorithm supports directed graphs in SAS/OR 14.1: http://support.sas.com/documentation/cdl/en/ornoaug/68159/HTML/default/viewer.htm#ornoaug_whatsnew_sect005.htm

RobPratt · ‎11-15-2015

For the NLP solver, you can use the MAX and ABS functions with variables as arguments. In that case, PROC OPTMODEL recognizes the nondifferentiability and uses numeric derivatives instead. But no automatic linearization is performed. If you just want to assign x4 to a value, outside of a solver call rather than as a constraint, the syntax you provided is perfectly legal.

RobPratt · ‎11-10-2015

Here's how you can write the variance (F) values to a data set. You can modify the code to get whatever additional values you want. PROC OPTMODEL; set <str> returns; read data contents into returns=[NAME]; VAR X{returns} >= 0; NUM COEFF{returns, returns}; read data covtable into [i=_NAME_] {j in returns} <coeff[i,j]=col(j)>; NUM R{returns}; read data meantable into {j in returns} <r[j]=col(j)>; num k; /* MINIMIZE THE VARIANCE OF THE PORTFOLIO’S TOTAL RETURN */ MINIMIZE F = SUM{I IN returns, J IN returns}COEFF[I,J]*X[I]*X[J]; /* SUBJECT TO THE FOLLOWING CONSTRAINTS */ CON BUDGET: SUM{I IN returns}X[I] = 1; CON GROWTH: SUM{I IN returns}R[I]*X[I] = 0.05*k; set KSET = 1..5; num Fsol {KSET}; do k=KSET; put k=; SOLVE; PRINT X; Fsol[k] = F.sol; end; create data soldata from [k] Fsol; quit;

Online Status	Offline
Date Last Visited	Friday