def __init__(self, num_datasets, dataset_size):
"""Creates a new GA parameter tuner.
This class can be used to tune the parameters of a GA but testing the
performance across a range of different parameter settings.
:param num_datasets: the number of datasets to generate
:param dataset_size: the size of the dataset to generate
"""
self._num_datasets = num_datasets
self._generator = TSPGenerator(dataset_size)
class GeneticAlgorithmParameterEstimation():
def __init__(self, num_datasets, dataset_size):
"""Creates a new GA parameter tuner.
This class can be used to tune the parameters of a GA but testing the
performance across a range of different parameter settings.
:param num_datasets: the number of datasets to generate
:param dataset_size: the size of the dataset to generate
"""
self._num_datasets = num_datasets
self._generator = TSPGenerator(dataset_size)
def perform_grid_search(self, params):
"""Perform a grid search over the range of parameters provided
This will create a single set of parameters for each of the ranges
specified in the params argument using scikit-learn's ParameterGrid
function.
Each set of sets of parameters is tested against multiple randomly
generated datasets (of the same size). The average fitness achieved
over all test datasets is taken as the measure of quality for the
parameters.
:param params: dictionary of ranges of parameters to be passed to ParameterGrid
:return: dictionary of the best parameter settings found
:rtype: dict
"""
param_grid = list(ParameterGrid(params))
self._param_fitness = []
datasets = [
self._generator.generate() for _ in range(self._num_datasets)
]
param_data = pd.DataFrame.from_dict(param_grid)
for setting in param_grid:
param_fitness = []
for dataset in datasets:
fitness = Simulator(**setting).score(dataset)
param_fitness.append(fitness)
median_fitness = np.median(param_fitness)
self._param_fitness.append(median_fitness)
param_data['fitness'] = self._param_fitness
return param_data
def get_best_fitness(self):
""" Get the average measure of fitness achieved over the datasets
:return: the average measure of fitness
:rtype: float
"""
return np.min(self._param_fitness)
class GeneticAlgorithmParameterEstimation():
def __init__(self, num_datasets, dataset_size):
"""Creates a new GA parameter tuner.
This class can be used to tune the parameters of a GA but testing the
performance across a range of different parameter settings.
:param num_datasets: the number of datasets to generate
:param dataset_size: the size of the dataset to generate
"""
self._num_datasets = num_datasets
self._generator = TSPGenerator(dataset_size)
def perform_grid_search(self, params):
"""Perform a grid search over the range of parameters provided
This will create a single set of parameters for each of the ranges
specified in the params argument using scikit-learn's ParameterGrid
function.
Each set of sets of parameters is tested against multiple randomly
generated datasets (of the same size). The average fitness achieved
over all test datasets is taken as the measure of quality for the
parameters.
:param params: dictionary of ranges of parameters to be passed to ParameterGrid
:return: dictionary of the best parameter settings found
:rtype: dict
"""
param_grid = list(ParameterGrid(params))
self._param_fitness = []
datasets = [self._generator.generate() for _ in range(self._num_datasets)]
param_data = pd.DataFrame.from_dict(param_grid)
for setting in param_grid:
param_fitness = []
for dataset in datasets:
fitness = Simulator(**setting).score(dataset)
param_fitness.append(fitness)
median_fitness = np.median(param_fitness)
self._param_fitness.append(median_fitness)
param_data['fitness'] = self._param_fitness
return param_data
def get_best_fitness(self):
""" Get the average measure of fitness achieved over the datasets
:return: the average measure of fitness
:rtype: float
"""
return np.min(self._param_fitness)
def __init__(self, num_datasets, dataset_size):
"""Creates a new GA parameter tuner.
This class can be used to tune the parameters of a GA but testing the
performance across a range of different parameter settings.
:param num_datasets: the number of datasets to generate
:param dataset_size: the size of the dataset to generate
"""
self._num_datasets = num_datasets
self._generator = TSPGenerator(dataset_size)
def generate(output_file, num_points):
generator = TSPGenerator(num_points)
logger.info("Generating %d points" % num_points)
data, _ = generator.generate()
save_dataset(data, output_file)
def generate_random_dataset(num_points):
logger.info('Generating %d uniformly random points' % num_points)
generator = TSPGenerator(num_points)
return generator.generate()
def generate(output_file, num_points):
generator = TSPGenerator(num_points)
logger.info("Generating %d points" % num_points)
data, _ = generator.generate()
save_dataset(data, output_file)
def generate_random_dataset(num_points):
logger.info('Generating %d uniformly random points' % num_points)
generator = TSPGenerator(num_points)
return generator.generate()
