Within the project description, there was a minimum set of low-level heuristics to be implemented, and optionally at least three other heuristics. In the text box below, list each additional low-level heuristic that you implemented on a new line and state the corresponding class name and heuristic type. If you reimplemented a required heuristic but used a different neighbourhood, for example reinsertion instead of swap, then you should state this. If you did not implement additional heuristics, then you should leave this field blank.

For example: 

Adjacent Swap (Mutation)

Given the following permutation representation of a solution to some instance of the Open-top Bus Routing problem using an integer array <0,1,2,3,4,5,6,7,8,9> determine for each of the low-level heuristics that were required to be implemented and listed below whether the candidate solution <0,1,4,3,2,5,6,7,8,9> is:

• Definitely reachable (Yes),
• Definitely not reachable (No), or
• May or may not be reachable as this would depend on unspecified details (Maybe), 

...after calling the HyFlex API method applyHeuristic(int iHeuristicIndex, int iCurrentSolutionIndex, int iCandidateSolutionIndex) with an intensity of mutation of 0.3 and a depth of search of 0.3.

Please indicate what type of device you used to run your hyper-heuristic for solving the unseen instance. (We will use this to ensure there are no inconsistencies between devices).

You should run your selection hyper-heuristic implemented as part of your project to solve an unseen instance of the Open-top Bus Routing problem.

You should have already downloaded the unseen instance as part of the "Files required for practical test" archive. Locate the file "unseen-instance.obr" inside the instances folder and import it into your project, updating your code as required by your design so you can run your hyper-heuristic on this new instance.

You should run your hyper-heuristic on the unseen instance using:

• The RANDOM initialisation method.
• A runtime of 5 nominal minutes as benchmarked - either as done yourself if using your own machine or using that given on Moodle if using a lab machine.
• A single trial where your experimental seed is set to 20260508L.

 After running your hyper-heuristic, report only the objective value of the best solution found below.

When designing and implementing your selection hyper-heuristic, you may have attempted to improve its performance by performing some form of experimentation. 

Briefly explain (in two or three sentences):

• Which parameter(s) or algorithmic component(s) you tried to fine-tune,
• State which method(s) you used to do the tuning/configuration, and 
• Explain anything else you did that goes beyond tuning your hyper-heuristic on all of the given instances.

If you did not attempt to improve the performance of the hyper-heuristic beyond your original implementation, then you must leave your answer blank. 

In the textbox below, clearly state which heuristic selection mechanism and which move acceptance method you used when designing your submitted learning-based selection hyper-heuristic, and then briefly explain in one or two sentences how your heuristic selection method "learns" about the low-level heuristics and then "selects" which heuristic(s) to apply.

Given the following Open-top Bus Routing Problem (OBR) instance, calculate and state the objective value of the solution encoded by the following permutation representation: <2,0,1,3,4>.

NAME : understanding_instance
COMMENT : instance for the exam
BUS_DEPOT_LOCATION
-2 0
POINTS_OF_INTEREST
0 1
-1 1
-1 0
0 0
1 0
EOF

Before continuing with the following exercises, confirm that running your implementation of Iterated Local Search with the default experimental configuration yields the following plots. 

By using your implementation of Iterated Local Search, and appropriately configuring the framework, find a configuration for iIntensityOfMutation (IOM) and iDepthOfSearch (DOS) such that Iterated Local Search performs better than DBHC alone when comparing their median performances for solving MAX-SAT instance 1 with a 10 second computational time budget.

For full context, here is the boxplot produced when running DBHC for 31 trials on MAX-SAT instance 1 given 10 seconds of runtime using the seed of -1414736884 ("hill climbing".hashcode()) from the previous lab exercise:

The results of the previous exercise may have surprised you.