Yet Another Math Programming Consultant

Your comment about (13) and (14) would also apply ...

2012-05-18T16:58:47.153-04:00

Your comment about (13) and (14) would also apply to (10) and (11), give or take a minus sign. That would further increase size and further reduce nonzeros. I wonder what impact it would have -- more in the same direction (Gurobi even better, CBC even worse), or not?

z<=x z<=y z>=x+y-1

2012-05-04T11:31:35.088-04:00

z<=x
z<=y
z>=x+y-1

What about binary AND? 0 + 0 = 0 0 + 1 = 0 1 +...

2012-05-04T10:48:34.606-04:00

What about binary AND?

0 + 0 = 0
0 + 1 = 0
1 + 0 = 0
1 + 1 = 1

Quick question: "all variables in all practic...

2012-05-03T12:19:37.109-04:00

Quick question:
"all variables in all practical models should have finite bounds"

Wouldn't this also compromise performance in some way (even if slightly)?

I am 100% behind Erwin on this one, a better analysis is valid.

I know that GAMS does a lot more; the comparison w...

2012-05-02T08:07:03.510-04:00

I know that GAMS does a lot more; the comparison was not meant to say that C++ is "better". Given that the C++-code is very simple and uses a very good permutation generation algorithm, I thought it would be interesting to compare to see what one pays for using GAMS.

As for solving the actual problem (which may very well be larger than 10 and thus out of reach of brute force

1. GAMS also creates a (sparse) data-structure and...

2012-05-02T05:31:24.754-04:00

1. GAMS also creates a (sparse) data-structure and stores all permutations (these are the sets p and allp in the example). So that is a little bit of extra work. But clearly: handwritten C/C++ can often beat a GAMS implementation.

2. When we solve the MIP formulation, we are even faster than C/C++ generating only the permutations.

As a comparison, on a Core i5 M480 (2.67 GHz) gene...

2012-05-02T01:46:05.704-04:00

As a comparison, on a Core i5 M480 (2.67 GHz) generating all permutations using the C++ built-in next_permutation takes about 0.5 seconds. The code is about as long as the GAMS code, but is much more readable.

Dear Erwin, Im a student. i try to solve a transpo...

2012-04-28T11:58:58.182-04:00

Dear Erwin,
Im a student. i try to solve a transpot model with gams. i used ur programs xls2gms too. but i cant to connect to code s. Can u send a sample code to me with xls2gms if u ok.
good working.
sory for bad english...
erselim@gmail.com

Sure, dual infeasiblity only makes sense for the c...

2012-04-23T06:37:26.703-04:00

Sure, dual infeasiblity only makes sense for the continuous case, I was talking about LP.

Assume you solving a MIP (the case of Erwin). Now ...

2012-04-23T05:33:22.880-04:00

Assume you solving a MIP (the case of Erwin). Now assume that the initial relaxation is unbounded. What can you concluded then? All you can conclude is that the problem is unbounded or infeasible. You would have to do a phase 1 problem to make a different conclusion. Most users would not like that I think.

So in my opinion there is a good reason why MIP codes report as they do.
<

I probably should add that, my previous comment no...

2012-04-20T10:53:55.452-04:00

I probably should add that, my previous comment notwithstanding, I do find the "infeasible or unbounded" message a PITA when helping other people with their models, since I typically have no intuition whether someone else's model is infeasible versus unbounded.

I've thumped the following drum before, but wh...

2012-04-20T10:51:37.545-04:00

I've thumped the following drum before, but what the heck, I'll do it again: all variables in all practical models should have finite bounds (and by "finite" I mean less than whatever huge value your solver equates with infinity). That way, the primal problem is never unbounded, so "infeasible or unbounded" will translate to "infeasible". If any primal

@Bo: Well said. Then again, one could make rerunni...

2012-04-20T02:08:51.600-04:00

@Bo: Well said. Then again, one could make rerunning the modified version the 'standard behavior' and give power users the option to turn it off. Not sure which is better...still thinking which I like better for the Optimization.Framework.

"Please don’t tell me “dual infeasible” is a ...

2012-04-19T16:34:31.326-04:00

"Please don’t tell me “dual infeasible” is a good return code."

I did not say it was good, but I definitely think it's better. And it's not because developers want the easiest way out.

"If you are working on a model you really want to know if the model is infeasible or unbounded"

I agree, but in many situations there is no way around

One of the reasons to not using a dual infeasible ...

2012-04-19T13:30:04.049-04:00

One of the reasons to not using a dual infeasible status is of course legacy, but also that many users don't know about duality and the farkas ray theory it leads to. Just like users want the SOLUTION and not the primal solution :-)

In my opinion it's because the optimizer shoul...

2012-04-19T13:26:09.313-04:00

In my opinion it's because the optimizer should not return "Unbounded". There should only be primal or dual infeasible return status, which is what we use in our software. The problem arise in situations where you don't know if the primal or dual feasible region is empty or not. For instance in presolve you can prove there exists no dual feasible solution, but to say it's

Since you are writing a wrapper. You might as well...

2012-04-06T17:23:27.048-04:00

Since you are writing a wrapper. You might as well do it in Python.

Python has much more powerful tools like (multiprocessing) which are easy to use and more readable.
http://docs.python.org/library/multiprocessing.html

Hi, Erwin. I´m working in a Vehicle Route Problem...

2012-04-06T13:26:43.428-04:00

Hi, Erwin.

I´m working in a Vehicle Route Problem wiht Gams, but i have problems with subtour elimination constraints because I don´t know how to write it in Gams. Would you like to help me. My problem have 4 vehicules, 12 cities and 1 depot.
My imputs are demand for each city, distance between each city and capacity for each vehicle that is exactly the same for each one.
<

Dear M. Kalvelagen, We thank you for pointing out...

2012-03-27T18:06:56.379-04:00

Dear M. Kalvelagen,

We thank you for pointing out LocalSolver on your blog. You have well understood the main principles behind LocalSolver:
1) multithreaded adaptive simulated annealing;
2) autonomous moves maintaining feasibility at each iteration (here is the main innovation);
3) highly-optimized incremental evaluation in order to explore millions of solutions per

Hi, Erwin!! I´m working on a vrp capacitated, and...

2012-03-25T23:56:59.997-04:00

Hi, Erwin!!

I´m working on a vrp capacitated, and i´m trying to add the aditional constraint to the model you shared, but i´ve had problems because I need to create a table with more than two dimentions that shows me capacity and cost at the same time. Do you have and example that could be helpful for me?

Thank in advance, i apreciate your help.

Juan Pablo Castro

I think I've seen this approach (or something ...

2012-03-22T16:44:38.445-04:00

I think I've seen this approach (or something closely related) under the name "de novo programming" (Milan Zeleny). Resource limits are sometimes stated as hard constraints when in fact that's just the amount that historically has been acquired.

Hi Erwin, I have a question. How to include severa...

2012-03-22T06:18:28.108-04:00

Hi Erwin, I have a question. How to include several uncertainty aspects into the model? For instance, in my farm model I want to do monte carlo simulation for irrigation water, price, and yield variability. Could you help how in this case loop command will look like?

is there any way to get that entire gms file?

2012-02-21T17:46:04.081-05:00

is there any way to get that entire gms file?

i'd please like to see more of that gams file,...

2012-02-21T17:44:10.250-05:00

i'd please like to see more of that gams file, i couldn't get which are tose optfile (1,2,3) and i supposse genbldint is the integer var for s scenario and yr year, am i right?

Interesting post, Erwin. I have two comments: (1) ...

2012-02-14T13:40:26.169-05:00

Interesting post, Erwin. I have two comments:
(1) Your comment about "it is often argued that a modeling language helps in maintaining models as fixes are easier to identify and implement than in a traditional programming language", is definitely true. But I think, quite a few solvers have API's that are pretty close to these advanced modeling languages and are quite readable.