Download Master CSV
Download Master CSV Version 2 Last Updated: April 18th, 14:00 Changes: Fixed percent-from-optimal-costs for grouse and osprey
Download Master CSV Version 3 Last Updated: April 30th, 15:52 Changes: Fixed entries for red and osprey; Fixed four entries of red; fixed entry for "red" in line 65
As in assignment 5, your code should figure out which solver would be best given the features of a problem instance. However, if you think it might not be the best, your code can hedge its bets and run multiple solvers.
The input is the same as in assignment 5 but the output is now a sequence of solvers as in:
robin,bluejay bluejay,sparrow,wrenIn the first case, the portfolio chose to run 2 solvers and 3 in the second case. We will simulate the portfolio by pretending that all solvers in a sequence are run for 5 minutes and the best solution is selected. Thus a portfolio with 3 solvers is assumed to take 15 minutes. The portfolio will be scored using the following function:
F = (.5 * % from best known) + (.5 * amount of time)where "amount of time" is either 5, 10 or 15 and "% from best known" will be allowed to vary depending on which solver and instance are chosen. The best performance of the solvers in the sequence will be used to determine % from best known. Note: for this function, lower is better.
The code should have the same input and signature as the portfolio from assignment 4. You can make this as complicated or simple as you want. For example, you can submit a slightly modified assignment 5 for this assignment and only select a single algorithm as was done in assignment 5. You could ignore the learned model and hard code a selection strategy based on your own analysis of the data. You could use multiple learned models. To paraphrase a quote from a favorite cooking show of mine "You will be judged on creativity and presentation." (see below) But getting something working that shows thought is the key criterion.