Multi objective Search - cyrille-artho/modbat GitHub Wiki

Modbat now supports a search-based approach relying on reinforcement learning for test case generation. The approach considers test case generation as an exploration versus exploration dilemma, and we address this dilemma by using a particular strategy of multi-armed bandits with multiple rewards.

To use this search-based approach, the user needs to provide values for 9 input parameters using flag options, including:

1. --bandit-tradeoff gives a value for the tradeoff of the bandit strategy (you can use 2 as a default value).
   
2. --backtrack-t-reward gives a reward value for the backtracked transition.
   
3. --self-t-reward gives a reward value for the self-transition.
   
4. --good-t-reward gives a reward value for the good/successful transition.
   
5. --fail-t-reward gives a reward value for the failed transition.
   
6. --precond-pass-reward gives a reward value for the passed precondition.
   
7. --precond-fail-reward gives a reward value for the failed precondition.
   
8. --assert-pass-reward gives a reward value for the passed assertion.
   
9. --assert-fail-reward gives a reward value for the failed assertion.
   Please note that all reward values should be in [0, 1].
   Then, to use the search-based approach, the user also needs to set the search mode by the flag option --search, which should be --search=heur.
   If you don't want to use the search-based approach, you can set --search=random, which enables random test case generation (default setting for Modbat), and you don't need to use reward flag options.

Example:

 modbat/build$ scala modbat.jar --classpath=modbat-examples.jar \
                                -n=1 -s=2455cfedeadbeef \
                                --no-redirect-out \
                                --dotify-path-coverage \
                                --search="heur" \
			        --bandit-tradeoff=2 \
	                        --backtrack-t-reward=0.8 \
                                --self-t-reward=0.4 \
		                --good-t-reward=0.6 \
			        --fail-t-reward=0.5 \
			        --precond-pass-reward=0.7 \
			        --precond-fail-reward=0.7 \
			        --assert-pass-reward=0.7 \
			        --assert-fail-reward=0.7 \
                                modbat.examples.SimpleModel

These rewards parameters can also be tuned to find the optimized settings by using multi-objective optimization approach. For the details, please check the reference.

Reference

1. Multi-objective Search for Model-based Testing