def testBestHillClimb(self):
"""Test the best hill climb algorithm.
Test whether it finds the best results as expected.
"""
# Initiate the build/test command and the log directory.
Task.InitLogCommand(None, None, 'output')
# Generate the testing specs.
specs = self._GenerateFlagSpecifications()
# Generate the initial generations for a test whose cost function is the
# summation of the values of all the flags.
generation_tasks = _GenerateAllFlagsTasks(specs)
generations = [
HillClimbingBestBranch(generation_tasks, set([]), specs)
]
# Test the algorithm. The cost function is the summation of all the values
# of all the flags. Therefore, the best value is supposed to be 0, i.e.,
# when all the flags are disabled.
_TestAlgorithm('sum(values[0:len(values)])', specs, generations, 0)
# This test uses a cost function that is the negative of the previous cost
# function. Therefore, the best result should be found in task with all the
# flags enabled.
cost_function = 'sys.maxint - sum(values[0:len(values)])'
all_flags = list(generation_tasks)[0].GetFlags()
cost = _ComputeCost(cost_function, specs, all_flags)
# Generate the initial generations.
generation_tasks = _GenerateNoFlagTask()
generations = [
HillClimbingBestBranch(generation_tasks, set([]), specs)
]
# Test the algorithm. The cost function is negative of the summation of all
# the values of all the flags. Therefore, the best value is supposed to be
# 0, i.e., when all the flags are disabled.
_TestAlgorithm(cost_function, specs, generations, cost)
def testIterativeElimination(self):
"""Test the iterative elimination algorithm.
Test whether it finds the best results as expected.
"""
# Initiate the build/test command and the log directory.
Task.InitLogCommand(None, None, 'output')
# Generate the testing specs.
specs = self._GenerateFlagSpecifications()
# Generate the initial generations. The generation contains the base line
# task that turns on all the flags and tasks that each turn off one of the
# flags.
generations = _GenerateAllIterativeEliminationTasks(specs)
# Test the algorithm. The cost function is the summation of all the values
# of all the flags. Therefore, the best value is supposed to be 0, i.e.,
# when all the flags are disabled.
_TestAlgorithm('sum(values[0:len(values)])', specs, generations, 0)
# This test uses a cost function that is the negative of the previous cost
# function. Therefore, the best result should be found in task with all the
# flags enabled.
all_flags_tasks = _GenerateAllFlagsTasks(specs)
cost_function = 'sys.maxint - sum(values[0:len(values)])'
# Compute the cost of the task that turns on all the flags.
all_flags = list(all_flags_tasks)[0].GetFlags()
cost = _ComputeCost(cost_function, specs, all_flags)
# Test the algorithm. The cost function is negative of the summation of all
# the values of all the flags. Therefore, the best value is supposed to be
# 0, i.e., when all the flags are disabled.
# The concrete type of the generation decides how the next generation will
# be generated.
_TestAlgorithm(cost_function, specs, generations, cost)
def testGeneticAlgorithm(self):
"""Test the Genetic Algorithm.
Do a functional testing here and see how well it scales.
"""
# Initiate the build/test command and the log directory.
Task.InitLogCommand(None, None, 'output')
# Generate the testing specs.
specs = self._GenerateFlagSpecifications()
# Initiate the build/test command and the log directory.
GAGeneration.InitMetaData(STOP_THRESHOLD, NUM_CHROMOSOMES, NUM_TRIALS,
specs, MUTATION_RATE)
# Generate the initial generations.
generation_tasks = GenerateRandomGATasks(specs, NUM_CHROMOSOMES,
NUM_TRIALS)
generations = [GAGeneration(generation_tasks, set([]), 0)]
# Test the algorithm.
_TestAlgorithm('sum(values[0:len(values)])', specs, generations, -1)
cost_func = 'sys.maxint - sum(values[0:len(values)])'
_TestAlgorithm(cost_func, specs, generations, -1)
def _ProcessGA(meta_data):
"""Set up the meta data for the genetic algorithm.
Args:
meta_data: the meta data for the genetic algorithm.
"""
assert BUILD_CMD in meta_data
build_cmd = meta_data[BUILD_CMD]
assert TEST_CMD in meta_data
test_cmd = meta_data[TEST_CMD]
if OUTPUT not in meta_data:
output_file = DEFAULT_OUTPUT
else:
output_file = meta_data[OUTPUT]
if CONF not in meta_data:
conf_file = DEFAULT_CONF
else:
conf_file = meta_data[CONF]
if NUM_BUILDER not in meta_data:
num_builders = DEFAULT_NUM_BUILDER
else:
num_builders = meta_data[NUM_BUILDER]
if NUM_TESTER not in meta_data:
num_testers = DEFAULT_NUM_TESTER
else:
num_testers = meta_data[NUM_TESTER]
if STOP_THRESHOLD not in meta_data:
stop_threshold = DEFAULT_STOP_THRESHOLD
else:
stop_threshold = meta_data[STOP_THRESHOLD]
if NUM_CHROMOSOMES not in meta_data:
num_chromosomes = DEFAULT_NUM_CHROMOSOMES
else:
num_chromosomes = meta_data[NUM_CHROMOSOMES]
if NUM_TRIALS not in meta_data:
num_trials = DEFAULT_NUM_TRIALS
else:
num_trials = meta_data[NUM_TRIALS]
if MUTATION_RATE not in meta_data:
mutation_rate = DEFAULT_MUTATION_RATE
else:
mutation_rate = meta_data[MUTATION_RATE]
specs = flags.ReadConf(conf_file)
# Initiate the build/test command and the log directory.
Task.InitLogCommand(build_cmd, test_cmd, output_file)
# Initiate the build/test command and the log directory.
GAGeneration.InitMetaData(stop_threshold, num_chromosomes, num_trials,
specs, mutation_rate)
# Generate the initial generations.
generation_tasks = testing_batch.GenerateRandomGATasks(
specs, num_chromosomes, num_trials)
generations = [GAGeneration(generation_tasks, set([]), 0)]
# Execute the experiment.
_StartExperiment(num_builders, num_testers, generations)
