Coursehero >> Pennsylvania >> Lehigh >> IE 418

Course Hero has millions of student submitted documents similar to the one below including study guides, homework solutions, papers, exam answer keys and textbook solutions.

Selection Variable Node Selection IE418: Integer Programming Je Linderoth Department of Industrial and Systems Engineering Lehigh University 2nd February 2005 Je Linderoth Variable Selection Node Selection IE418 Integer Programming Boring Stu Extra Linux Class: 8AM 11AM, Wednesday February 9. Room ??? Accounts and Passwords http://coral.ie.lehigh.edu has user information Please use ipxx account when solving your IP problems in COR@L! Use yppasswd to change your password Homework: Due 2/9. (One week warning) Make a copy of your answers before you hand them in A Even better, use LTEXto write up your answers! A special present for you! Je Linderoth IE418 Integer Programming Variable Selection Node Selection Please don t call on me! Name some keys to solving integer programs Relaxations Formulation Formulation so the Relaxation is good Small M s good. Big M s baaaaaaaaaaaaaaaaaaaad....... How does branch-and-bound work? Je Linderoth Variable Selection Node Selection IE418 Integer Programming LP-based Branch and Bound Algorithm 1 To start, derive an lower bound L using a heuristic method (if possible). Put the original problem on the candidate list. Select a problem S from the candidate list and solve the LP relaxation to obtain the bound u(S) If the LP is infeasible node can be pruned. Otherwise, if u(S) L node can be pruned. Otherwise, if u(S) > L and the solution is feasible for the MILP set L u(S). Otherwise, branch. Add the new subproblems to the list. 2 3 4 If the candidate list in nonempty, go to Step 2. Otherwise, the algorithm is completed. Je Linderoth IE418 Integer Programming Variable Selection Node Selection Let s Do An Example maximize z = 5x1 + 4x2 + x3 + 7x4 subject to x1 + x2 5 x3 + x4 3 x1 x3 + x4 16 10x1 + 6x2 45 x1 , x2 0 x1 , x2 , x3 , x4 Z Je Linderoth Variable Selection Node Selection IE418 Integer Programming Solving the Example with B&B Your picture(s) here... Je Linderoth IE418 Integer Programming Variable Selection Node Selection Why? Strong Branching Pseudo Costs Branching Finale Priorities and SOS The Goal of Branching We want to divide the current problem into two or more subproblems that are easier than the original. We would like to choose the branching that minimizes the sum of the solution times of all the created subproblems. This is the solution of the entire subtree rooted at the node. How do we know how long it will take to solve each subproblem? Answer: We don t. Idea: Try to predict the di culty of a subproblem. Je Linderoth Variable Selection Node Selection IE418 Integer Programming Why? Strong Branching Pseudo Costs Branching Finale Priorities and SOS A Good Branching Imagine that when I branch, the value of the linear programming relaxation changes a lot! I can prune the node, or should be able to prune it quickly So, for a given potential branching, I would like to know the upper bound that would result from processing each subproblem. The branching that changes these bounds the most is the best branching. Be Creative! What are some ideas you have for deciding on a branching variable? Je Linderoth IE418 Integer Programming Variable Selection Node Selection Why? Strong Branching Pseudo Costs Branching Finale Priorities and SOS Predicting the Di culty of a Subproblem How can I (quickly?) estimate the upper bounds that would result? Partially solve the LP relaxation in each of the subproblems by performing a given number of dual simplex pivots. Since we are using dual simplex, this gives us a valid bound. Why? This technique is usually called strong branching. A cheaper alternative is to use pseudo-costs. Je Linderoth Variable Selection Node Selection IE418 Integer Programming Why? Strong Branching Pseudo Costs Branching Finale Priorities and SOS Strong Branching Details In the case of strong branching, it may be too expensive to evaluate all possible candidates for branching. How do we choose the candidates to evaluate? We choose them based on an estimate of their e ectiveness that is very cheap to evaluate. One method is to choose inequalities whose left hand side is furthest from being an integer For 0-1 variables, this means those whose values are closest to 0.5. We might also account for the size of the objective function coe cient. Je Linderoth IE418 Integer Programming Variable Selection Node Selection Why? Strong Branching Pseudo Costs Branching Finale Priorities and SOS Strong Branching Details The number of candidates to evaluate must be determined empirically. E ective branching is more important near the top of the tree. We might want to evaluate more candidates near the top of the tree. More candidates almost always results in smaller trees, but the expense eventually causes an increase in running time. How many dual simplex pivots should we do? Je Linderoth Variable Selection Node Selection IE418 Integer Programming Why? Strong Branching Pseudo Costs Branching Finale Priorities and SOS Using Pseudo Costs The pseudo-cost of a variable is an estimate of the per-unit change in the objective function from forcing the value of the variable to be rounded up or down. Like a gradient! For each variable xj , we maintain an up and a down pseudo-cost, denoted Pj+ and Pj . Let fj be the current (fractional) value of variable xj . An estimate of the change in objective function in each of the subproblems resulting from branching on xj is given by + Dj = Pj+ (1 fj ), Dj = Pj fj . The question is how to get the pseudo-costs. Je Linderoth IE418 Integer Programming Variable Selection Node Selection Why? Strong Branching Pseudo Costs Branching Finale Priorities and SOS Obtaining and Updating Pseudo Costs Typically, the pseudo-costs are obtained from empirical data. We observe the actual change that occurs after branching on each one of the variables and use that as the pseudo-cost. We can either choose to update the pseudo-cost as the calculation progresses or just use the rst pseudo-cost found. Several authors have noted that the pseudo-costs tend to remain fairly constant. The only remaining question is how to initialize. Possibilities: Use the objective function coe cient. Use the of average all known pseudo-costs. Explicity initialize the pseudocosts using strong branching Je Linderoth Variable Selection Node Selection IE418 Integer Programming Why? Strong Branching Pseudo Costs Branching Finale Priorities and SOS What Does The Most Mean If we are doing typical variable branching, we create two children and have estimates of the amount the bound will change for each child How do we combine the two nunbers together to form one measure of goodness for a potential branch? Suggest to branch on the variable + + j = arg max{ 1 min{Dj , Dj } + 2 max{Dj , Dj }. 2 = 0 we want to maximize the minimum degradation on the branch ( 1 , 2 ) = (2, 1) seems pretty good Je Linderoth IE418 Integer Programming Variable Selection Node Selection Why? Strong Branching Pseudo Costs Branching Finale Priorities and SOS Putting it All Together Here are the choices we ve discussed in branching: Should we use strong branching or pseudo-costs? Pseudo-costs How should we initialize? How should we update? Strong branching How do we choose the list of branching candidates? How many pivots to do on each? Once we have the bound estimates, how do we choose the nal branching? Ultimately, we must use empirical evidence and intuition to answer these questions. Je Linderoth Variable Selection Node Selection IE418 Integer Programming Why? Strong Branching Pseudo Costs Branching Finale Priorities and SOS Priorities How Much Do You Know? You are smarter than integer programming! If you have problem speci c knowledge, use it to determine which variable to branch ong Branch on the important variables rst First decide which warehouses to open, then decide the vehicle routing Branch on earlier (time-based) decisions rst. There are mechanisms for giving the variables a priority order, so that if two variables are fractional, the one with the high priority is branched on rst Or, rst branch on all these variables before you branch on the next class, etc. Je Linderoth IE418 Integer Programming Variable Selection Node Selection Why? Strong Branching Pseudo Costs Branching Finale Priorities and SOS GUB/SOS1 Branching xj {0, 1} j 10000 xj = 1 j=1 Which branching do you think would be better? 1 2 x1 = 1 & x1 = 0( 500 j=1 xj =1& 10000 j=2 = 1), 10000 j=501 xj = 1 or The answer is It depends But the answer is almost assuredly (2). It is probably even better to look at the (infeasible) LP relaxation and put 1/2 on each side (Just don t break it in the middle). IE418 Integer Programming Je Linderoth Variable Selection Node Selection Why? Strong Branching Pseudo Costs Branching Finale Priorities and SOS SOS2 Branching { 1 , 2 , ... 100 } is an SOS2 Suppose: 1 = 0.2 6 = 0.1 8 = 0.3 10 = 0.1 17 = 0.05 99 = 0.25 If k > 0, then feasible solutions have 1 = = k 1 = 0, or k+1 = = 100 = 0 Pk 1 j=1 j = 0 Pn j=k+1 j = 0 The $64 Question How would you branch? Plus the (infeasible) point is excluded on both branches. Je Linderoth IE418 Integer Programming Variable Selection Node Selection Why? Best First Depth-First Best Estimate Choices in Branch and Bound Node Selection Another important parameter to consider in branch and bound is the strategy for selecting the next subproblem to be processed. In choosing a search strategy, we might consider two di erent goals: Minimizing overall solution time. Finding a good feasible solution quickly. Je Linderoth Variable Selection Node Selection IE418 Integer Programming Why? Best First Depth-First Best Estimate The Best First Approach One way to minimize overall solution time is to try to minimize the size of the search tree. We can achieve this choose the subproblem with the best bound (highest upper bound if we are maximizing). A candidate node is said to be critical if its bound exceeds the value of an optimal solution solution to the IP. Every critical node will be processed no matter what the search order. Best rst is guaranteed to examine only critical nodes, thereby minimizing the size of the search tree. Je Linderoth IE418 Integer Programming Variable Selection Node Selection Why? Best First Depth-First Best Estimate Drawbacks of Best First Doesn t necessarily nd feasible solutions quickly Feasible solutions are more likely to be found deep in the tree Node setup costs The linear program being solved may change quite a bit more one iteration to the next Memory usage. It can require a lot of memory to store the candidate list Je Linderoth Variable Selection Node Selection IE418 Integer Programming Why? Best First Depth-First Best Estimate The Depth First Approach The depth rst approach is to always choose the deepest node to process next. Just dive until you prune, then back up and go the other way This avoids most of the problems with best rst: The number of candidate nodes is minimized (saving memory). The node set-up costs are minimized LPs change very little from one iteration to the next Feasible solutions are usually found quickly Unfortunately, if the initial lower bound is not very good, then we may end up processing lots of non-critical nodes. We want to avoid this extra expense if possible. Je Linderoth IE418 Integer Programming Variable Selection Node Selection Why? Best First Depth-First Best Estimate Estimate-based Strategies: Finding Feasible Solutions Let s focus on a strategy for nding feasible solutions quickly. One approach is to try to estimate the value of the optimal solution to each subproblem and pick the best. For any subproblem Si , let si = j min(fj , 1 fj ) be the sum of the integer infeasibilities, i zU be the upper bound, and zL the global lower bound. Also, let S0 be the root subproblem. The best projection criterion is Ei = i zU + 0 zL zU s0 si The best estimate criterion uses the pseudo-costs to obtain i Ei = zU + j min Pj fj , Pj+ (1 fj ) Je Linderoth Variable Selection Node Selection IE418 Integer Programming Why? Best First Depth-First Best Estimate Next Time: You should read (as review, and for more information) N&W II.4.1, II.4.2 Introduction to IP software Who knows what an MPS le is? Je Linderoth IE418 Integer Programming

Find millions of documents here - Study Guides, Homework Solutions, Papers, Exam Answer Keys and more. Course Hero has millions of course related materials that will enable you to learn better, faster and get an A in all your courses.
Below is a small sample set of documents: