def cs(cls, **kwargs):
"""
Each tree represents a set of hyperparameter spaces and is a, possibly
infinite, set of configurations.
Parameters
----------
data
If given, 'data' limits the search interval of some hyperparameters
Returns
-------
Tree representing all the possible hyperparameter spaces.
"""
tree = cls.cs_impl(**kwargs)
if 'config_spaces' in kwargs:
del kwargs['config_spaces']
hps = tree.hps.copy()
hps['module'] = CatHP(choice, a=[cls.__module__])
hps['class'] = CatHP(choice, a=[cls.__name__])
# Freeze args passed via kwargs
for k, v in kwargs.items():
hps[k] = FixedHP(v)
return tree.updated(hps=hps)
def cs_impl(cls):
hps = {
'n_neighbors': IntHP(uniform, low=1, high=15),
'weights': CatHP(choice, a=['uniform', 'distance']),
'algorithm': CatHP(choice, a=['kd_tree', 'ball_tree']),
'leaf_size': IntHP(uniform, low=15, high=100),
'p': IntHP(uniform, low=1, high=5)
} #default metric: 'minkowski'
return ConfigSpace(name='KNN', hps=hps)
def cs_impl(cls, config_spaces):
hps = [
CatHP('configs',
cls.sampling_function,
config_spaces=config_spaces[0]),
CatHP('reduce',
cls.sampling_function,
config_spaces=config_spaces[1])
]
return ConfigSpace(name=cls.__name__, hps=hps)
def cs_impl(cls):
hps = {
'n_neighbors': IntHP(uniform, low=2, high=1000),
'weights': CatHP(choice, a=['uniform', 'distance']),
'algorithm': CatHP(choice,
a=['auto', 'ball_tree', 'kd_tree', 'brute']),
'leaf_size': IntHP(uniform, low=2, high=1000),
'p': IntHP(uniform, low=1, high=6)
}
return ConfigSpace(name=cls.__name__, hps=hps)
def cs_impl(cls):
hps = {
'C': RealHP(uniform, low=0.0, high=10000.0),
'kernel': CatHP(choice, a=['linear', 'poly', 'rbf', 'sigmoid']),
'degree': IntHP(uniform, low=0, high=1000),
'gamma': CatHP(choice, a=['scale', 'auto']),
'coef0': RealHP(uniform, low=0.0, high=1000.0),
'shrinking': CatHP(choice, a=[True, False]),
'probability': CatHP(choice, a=[True, False]),
'tol': RealHP(uniform, low=1e-6, high=0.5),
'max_iter': IntHP(uniform, low=-1, max=10000),
'decision_function_shape': CatHP(choice, a=['ovo', 'ovr'])
}
return ConfigSpace(name=cls.__name__, hps=hps)
def cs_impl(cls, data=None):
hps = {
'sampling_strategy':
CatHP(choice,
items=['majority', 'not minority', 'not majority', 'all'])
}
return ConfigSpace(name=cls.__name__, hps=hps)
def cs_impl(cls):
# Sw
# cs = ConfigSpace('Switch')
# st = cs.start()
# st.add_children([a.start, b.start, c.start])
# cs.finish([a.end,b.end,c.end])
hps = {
'criterion': CatHP(choice, a=['gini', 'entropy']),
'splitter': CatHP(choice, a=['best']),
'class_weight': CatHP(choice, a=[None, 'balanced']),
'max_features': CatHP(choice, a=['auto', 'sqrt', 'log2', None]),
'max_depth': IntHP(uniform, low=2, high=1000),
'min_samples_split': RealHP(uniform, low=1e-6, high=0.3),
'min_samples_leaf': RealHP(uniform, low=1e-6, high=0.3),
'min_weight_fraction_leaf': RealHP(uniform, low=0.0, high=0.3),
'min_impurity_decrease': RealHP(uniform, low=0.0, high=0.2)
}
return ConfigSpace(name='DT', hps=hps)
def cs_impl(cls):
hps = {
'sampling_strategy':
CatHP(choice,
items=[
'not minority', 'not majority', 'all', 'minority', 'auto'
]),
'k_neighbours':
IntHP(uniform, low=2, high=1000)
}
return ConfigSpace(name=cls.__name__, hps=hps)
def cs_impl(cls):
hps = {
'n_estimators': IntHP(uniform, low=2, high=1000),
'criterion': CatHP(choice, a=['mse', 'mae', 'friedman_mse']),
'max_features': CatHP(choice, a=['auto', 'sqrt', 'log2', None]),
'max_depth': IntHP(uniform, low=2, high=100),
'min_samples_split': RealHP(uniform, low=1e-6, high=0.5),
'min_samples_leaf': RealHP(uniform, low=1e-6, high=0.5),
'min_weight_fraction_leaf': RealHP(uniform, low=0.0, high=0.5),
'min_impurity_decrease': RealHP(uniform, low=0.0, high=0.2),
'loss': CatHP(choice, a=['ls', 'lad', 'huber', 'quantile']),
'learning_rate': RealHP(uniform, low=1e-4, high=1e-1),
'subsample': RealHP(uniform, low=0.1, high=0.9),
'alpha': RealHP(uniform, low=0.1,
high=0.9), #ignored if loss != huber or quantile
'validation_fraction': RealHP(uniform, low=0.1,
high=0.9), # Using Early Stopping #
'n_iter_no_change': IntHP(uniform, low=5,
high=20) # #
}
return ConfigSpace(name='GB', hps=hps)
def cs_impl(cls):
hps = {'function': CatHP(choice, items=['mean'])}
return ConfigSpace(name=cls.__name__, hps=hps)
def cs_impl(cls, config_spaces):
hps = {'configs': CatHP(cls.sampling_function,
config_spaces=config_spaces)}
return ConfigSpace(name=cls.__name__, hps=hps)
def cs_impl(cls, data=None):
hps = {
'@with_mean/std':
CatHP(choice, items=[(True, False), (False, True), (True, True)])
}
return ConfigSpace(name=cls.__name__, hps=hps)
def cs_impl(cls):
hps = {'@nb_type': CatHP(choice, items=['GaussianNB', 'BernoulliNB'])}
return ConfigSpace(name='NB', hps=hps)
def cs_impl(cls):
# TODO: cirar funcao no data
hps = {'field': CatHP(choice, items=['X', 'Y', 'Z'])}
return ConfigSpace(name=cls.__name__, hps=hps)
def cs_impl(self):
hps = {
'@nb_type': CatHP(choice, items=["MultinomialNB", "ComplementNB"])
}
return ConfigSpace(name='NB', hps=hps)
def cs_impl(cls):
# TODO: cirar funcao no data
hps = [CatHP('field', choice, itens=['X', 'Y', 'Z'])]
return ConfigSpace(name=cls.__name__, hps=hps)
def cs_impl(cls):
hps = {'feature_range': CatHP(choice, items=[(-1, 1), (0, 1)])}
return ConfigSpace(name=cls.__name__, hps=hps)