The grammatical evolution algorithms build on top of the standard genetic algorithm. A key difference is that for the grammatical evolution algorithms the Android test cases are not created and executed when updating the population (i.e. creating the initial population, doing cross over and mutation). Instead, the algorithms only produce and alter codon sequences (i.e. lists of integers) in those phases.
The mapping from the codon sequences to Android test cases happens during fitness evaluation. For this we employ the GenotypePhenotypeMappedFitnessFunction. It is setup with a method to convert the codon sequence to an Android test case and a fitness function that can measure the fitness of the resulting Android test case. The latter is the same fitness function that would be used for the standard genetic algorithm. The method to convert the codon sequence to an Android test case depends on which grammatical evolution approach is chosen. For the standard grammatical evolution every codon with an odd position (1, 3, 5, ...) determines whether the Android test case it produces should end or not. Every codon at an even (2, 4, 6, ...) position determines which of the available actions will be performed. A test case ends either when the first codon at an odd position with a test case ending codon value is encountered or when an action results in the app being closed. After converting all chromosomes to test cases they are evaluated with the given fitness function.
Cross over and mutation functions are performed on the codon sequences. Currently there is only one implementation of a cross over function: the IntegerSequencePointCrossOverFunction. For mutation there are two options. The IntegerSequencePointMutationFunction which randomly changes a single codon of the sequence. The IntegerSequenceLengthMutationFunction randomly removes or adds codons to the sequence. This mutation function is used for the grammatical evolution algorithm that uses the length of the codon sequence as the length of the resulting Android test case.
Creating the initial population is very simple as we only need to produce lists with random integers. This is done by the IntegerSequenceChromosomeFactory.