introduction to mosel and xpress
play

Introduction to Mosel and Xpress ORLAB - Operations Research - PowerPoint PPT Presentation

Introduction to Mosel and Xpress ORLAB - Operations Research Laboratory Stefano Gualandi Politecnico di Milano, Italy April 15, 2011 Stefano Gualandi Introduction to Mosel and Xpress From Modeling to Strategies Model Algorithms Solution (s)


  1. Introduction to Mosel and Xpress ORLAB - Operations Research Laboratory Stefano Gualandi Politecnico di Milano, Italy April 15, 2011 Stefano Gualandi Introduction to Mosel and Xpress

  2. From Modeling to Strategies Model Algorithms Solution (s) Modeling Interpretation Problem Strategy Stefano Gualandi Introduction to Mosel and Xpress

  3. From Modeling to Strategies Data Algorithms Solution (s) Math Model Modeling Problem Stefano Gualandi Introduction to Mosel and Xpress

  4. From Modeling to Strategies Data Algorithms Solution (s) Math Model Modeling Problem Stefano Gualandi Introduction to Mosel and Xpress

  5. From Modeling to Strategies Data Algorithms Solution (s) Solver Math Model Modeling Problem Stefano Gualandi Introduction to Mosel and Xpress

  6. From Modeling to Strategies Data - Tables Solver - Charts Math Model Modeling Problem Stefano Gualandi Introduction to Mosel and Xpress

  7. From Modeling to Strategies Data - Tables Solver - Charts Math Model Modeling Problem Stefano Gualandi Introduction to Mosel and Xpress

  8. Why lab sessions are useful? Lab sessions close the gap between theory and practice. The goals are: ◮ To train intuition and skills necessary to formulate models ◮ To learn algebraic modeling languages ◮ To understand functionalities, strength, and weakness of solvers ◮ To acquire skills in interpreting solutions Warning: Modeling is more an art than a science Stefano Gualandi Introduction to Mosel and Xpress

  9. Model formulation ◮ Although modeling is a kind of art , good practices exist: Do not reinvent the wheel ◮ Both academia and industry have developed models for a wide classes of problems: ◮ Set covering ◮ Set partitioning ◮ Matching ◮ Routing ◮ Scheduling ◮ (in Xpress, see the menu “Wizard-¿Complete models”) ◮ Different application domain problems do have the same mathematical model Hint : study by heart all the examples of the course notes! Warning: Lab sessions complement exercise sessions Stefano Gualandi Introduction to Mosel and Xpress

  10. Algebraic modeling languages ◮ Algebraic modeling languages have a well-defined syntax and semantic close to mathematics ◮ They are declarative, that is, they say what to solve (the model), but not how (you do not implement algorithms) ◮ They force users to distinguish between the logic structure of the problem (i.e., the model) and the numerical data (i.e., the instances of the problem) ◮ During the lab sessions we use Mosel . WARNING : read carefully the error messages! Stefano Gualandi Introduction to Mosel and Xpress

  11. Solvers ◮ A solver is a software that takes as input a problem description (i.e., model+data) and output a solution (if one exists) ◮ Depending on the problem structure different solvers should be used: ◮ Linear Programming solver ◮ Integer Programming solver ◮ Semi-definite Programming solver ◮ Non-Linear solver ◮ Solvers have many options which can improve the resolution time if properly used ◮ During the lab sessions you will solve mainly Linear and Integer programs (“mmxprs” module of Mosel). Stefano Gualandi Introduction to Mosel and Xpress

  12. Interpreting solution ◮ Solvers give correct solutions with respect to your model... but is the given solution valid for the given problem? ◮ If they are not valid, 99.999% of the times you wrote the wrong model or the wrong data (50% of the times you write syntax errors!) ◮ Common pitfalls: ◮ unbounded variables ◮ missing constraints ◮ missunderstood constraints ◮ wrong quantification in the summation of constraints ◮ wrong objective function ◮ You should know what solutions look like ◮ Getting a solution is not the end of the story! Stefano Gualandi Introduction to Mosel and Xpress

  13. What you do during lab sessions Data - Tables Solver - Charts Math Model Modeling Interpretation Problem Strategy Stefano Gualandi Introduction to Mosel and Xpress

  14. What you do during lab sessions FILE.dat MOSEL - writeln Xpress - Excel FILE.mos Modeling Interpretation Problem Strategy Stefano Gualandi Introduction to Mosel and Xpress

  15. A short introduction to Mosel In Mosel , models and data are written in two separate text files: ◮ FILENAME .mos : it contains the model of the problem, i.e., variables, constraints, and objective function (plus basic scripting commands) ◮ FILENAME .dat : it contains the data specifing an instance of the problem, i.e., the coefficients of the constraints and of the objective function Mosel has itw own IDE, debugger, examples and MANUALS. Stefano Gualandi Introduction to Mosel and Xpress

  16. Mosel blocks model MODELNAME uses "mmxprs","mmive"; declarations [...] end-declarations initializations from ’FILE.dat’ [...] end-initializations declarations [...] end-declarations Stefano Gualandi Introduction to Mosel and Xpress

  17. Mosel keywords ◮ range : defines set of integer (used for indexing) ◮ set of : defines set of elements (e.g., range is a set of integers) ◮ array : defines constant parameters like constants, vectors, or matrices ◮ mpvar : defines continuous variables. They can be defined also as ◮ is free ◮ is integer ◮ is binary ◮ sum : defines a linear sum ◮ lincrt : defines a family of constraints ◮ minimize or maximize : defines the objective function Stefano Gualandi Introduction to Mosel and Xpress

  18. Exercise 1: the math model Given: ◮ A matrix m × n ◮ I = { 1 , . . . , n } indexes of columns of A ◮ J = { 1 , . . . , m } indexes of rows of A Stefano Gualandi Introduction to Mosel and Xpress

  19. Exercise 1: the math model Given: ◮ A matrix m × n ◮ I = { 1 , . . . , n } indexes of columns of A ◮ J = { 1 , . . . , m } indexes of rows of A Solve the minimization problem: � min c i x i i ∈ I s.t. Ax ≤ b , x ∈ R n . x ≥ 0 , Stefano Gualandi Introduction to Mosel and Xpress

  20. Exercise 1: ex-1.mos model ExampleOne uses "mmxprs", "mmive" declarations I : range ! set of integers (indices) J : range ! set of integers (indices) c : array(I) of integer ! cost vector b : array(J) of real ! rhs A : array(J,I) of real ! coef. matrix x : array(I) of mpvar myCst: array(J) of linctr end-declarations ! DATA DEFINITION initializations from ’ex-1.dat’ I J c b A end-initializations Stefano Gualandi Introduction to Mosel and Xpress

  21. Exercise 1: ex-1.mos ! Objective function Cost := sum (i in I) c(i)*x(i) ! Linear constraints forall (j in J) myCst(j) := sum (i in I) A(j,i)*x(i) <= b(j) ! Make variable(s) integer forall(i in I) x(i) is_integer writeln("Begin running model") minimize(Cost) writeln("End running model") Stefano Gualandi Introduction to Mosel and Xpress

  22. Exercise 1: ex-1.dat I : [1 2 3] J : [1 2] c : [2 4 3] b : [-3 9.2] A : [-1 2 3.1 3.3 -7 -1] Stefano Gualandi Introduction to Mosel and Xpress

  23. A tour with Mosel and Xpress Exercises: ◮ Download the ex-1.mod and ex-1.dat files and solve the problem. What is the optimal solution? ◮ If the variable x were continuous, the optimum would be greater or smaller? Why? ◮ And what if the x variable were binary ? The model/data are still correct? Why? Stefano Gualandi Introduction to Mosel and Xpress

  24. Exercise 2: Mobile phones, problem statement We have an assemblage kit containing pieces of different colors and dimensions, each corresponding to a specific component of a mobile phone. We have: 10 display modules; 18 memory modules; 12 transmission modules; 6 keyboards; 9 mouse; 10 microcameras. Appropriately combining these elements, we can assemble mobile phones of two different models. Each phone of type A brings 3 points; each phone of type B brings 8 points. The game consists in assembling the pieces that we have, trying to make the highest possible score. Stefano Gualandi Introduction to Mosel and Xpress

  25. Mobile phones: modeling ◮ Sets: ◮ T = set of types of mobile phones ◮ C = set of components ◮ Parameters: ◮ p i = score for model i ◮ a ij = number of components j necessary for a type i ◮ d j = availability of component j ◮ Variables: ◮ x i = number of produced mobile phone of type i Stefano Gualandi Introduction to Mosel and Xpress

  26. Mobile phones: modeling Objective function: � max p i x i i ∈ M Constraints: � a ij x i ≤ d j , ∀ j ∈ C , i ∈ M x i ≥ 0 , integer , ∀ i ∈ M . Stefano Gualandi Introduction to Mosel and Xpress

  27. Mobile phones: questions 1. Download the ex-2.dat file. 2. Download the ex-2.mos file and complete the model. Solve the problem. What is the optimal solution? 3. load the model and data file 4. solve the problem: which is the optimum? 5. which and how many mobiles are produced? 6. what if you had only 0 mouse component? Try to modify the .dat file and explain what happen. 7. what if you had 8 keyboards component? Try to modify the .dat file and explain what happen. 8. what if you had 9 keyboards component? Try to modify the .dat file and explain what happen. Stefano Gualandi Introduction to Mosel and Xpress

Recommend


More recommend