def _cs_impl(cls) -> CS:
params = {
"path": FixedP("./"),
"name": FixedP("blalbairis.arff"),
"matrices_hash": FixedP("1234567890123456789"),
}
return CS(nodes=[Node(params=params)])
def _cs_impl(cls):
params = {
'path': FixedP('./'),
'name': FixedP('iris.arff'),
'description': FixedP('No description.'),
'matrices_hash': FixedP('1234567890123456789')
}
return TransformerCS(Node(params=params))
def _cs_impl(cls) -> CS:
# todo: set random seed; set 'cache_size'
param = {
"n": RealP(uniform, low=0.0, high=1.0),
"copy": FixedP(True),
"whiten": FixedP(False),
"svd_solver": FixedP("auto"),
"tol": FixedP(0.0),
"iterated_power": FixedP("auto"),
}
return CS(nodes=[Node(param)])
def _cs_impl(cls):
# todo: set random seed; set 'cache_size'
kernel_linear = Node({"kernel": FixedP("linear")})
kernel_poly = Node(
{
"kernel": FixedP("poly"),
"degree": IntP(uniform, low=0, high=10),
"coef0": RealP(uniform, low=0.0, high=100),
}
)
kernel_rbf = Node({"kernel": FixedP("rbf")})
kernel_sigmoid = Node({"kernel": FixedP("sigmoid"), "coef0": RealP(uniform, low=0.0, high=100),})
kernel_nonlinear = Node(
{"gamma": RealP(uniform, low=0.00001, high=100)}, children=[kernel_poly, kernel_rbf, kernel_sigmoid],
)
top = Node(
{
"C": RealP(uniform, low=1e-4, high=100),
"shrinking": CatP(choice, items=[True, False]),
"probability": FixedP(False),
"tol": OrdP(choice, items=[0.000001, 0.00001, 0.0001, 0.001, 0.01, 0.1, 1, 10, 100, 1000, 10000,],),
"class_weight": CatP(choice, items=[None, "balanced"]),
# 'verbose': [False],
"max_iter": FixedP(1000000),
"decision_function_shape": CatP(choice, items=["ovr", "ovo"]),
},
children=[kernel_linear, kernel_nonlinear],
)
return CS(nodes=[top])
def _cs_impl(cls):
params = {
"criterion": CatP(choice, items=["gini", "entropy"]),
"splitter": FixedP("best"),
"class_weight": CatP(choice, items=[None, "balanced"]),
"max_features": CatP(choice, items=["auto", "sqrt", "log2", None]),
"max_depth": IntP(uniform, low=2, high=1000),
"min_samples_split": RealP(uniform, low=1e-6, high=0.3),
"min_samples_leaf": RealP(uniform, low=1e-6, high=0.3),
"min_weight_fraction_leaf": RealP(uniform, low=0.0, high=0.3),
"min_impurity_decrease": RealP(uniform, low=0.0, high=0.2),
}
return CS(nodes=[Node(params=params)])
def _cs_impl(cls):
params = {
'criterion': CatP(choice, items=['gini', 'entropy']),
'splitter': FixedP('best'),
'class_weight': CatP(choice, items=[None, 'balanced']),
'max_features': CatP(choice, items=['auto', 'sqrt', 'log2', None]),
'max_depth': IntP(uniform, low=2, high=1000),
'min_samples_split': RealP(uniform, low=1e-6, high=0.3),
'min_samples_leaf': RealP(uniform, low=1e-6, high=0.3),
'min_weight_fraction_leaf': RealP(uniform, low=0.0, high=0.3),
'min_impurity_decrease': RealP(uniform, low=0.0, high=0.2)
}
return TransformerCS(nodes=[Node(params=params)])
def hold(cs, **kwargs):
"""Freeze args passed via kwargs. Only applicable to ComponentCS.
Keyworded args are used to freeze some parameters of
the algorithm, regardless of what a CS sampling could have chosen.
TODO: it may be improved to effectively traverse and change the tree
in-place, not just extend overwritting it
"""
cs = cs.cs
new_nodes = []
for node in cs.nodes:
params = {} if node.params is None else node.params.copy()
for k, v in kwargs.items():
params[k] = FixedP(v)
new_nodes.append(node.updated(params=params))
return cs.updated(nodes=new_nodes)
def _cs_impl(cls):
# todo: set random seed; set 'cache_size'
kernel_linear = Node({'kernel': FixedP('linear')})
kernel_poly = Node({
'kernel': FixedP('poly'),
'degree': IntP(uniform, low=0, high=10),
'coef0': RealP(uniform, low=0.0, high=100)
})
kernel_rbf = Node({'kernel': FixedP('rbf')})
kernel_sigmoid = Node({
'kernel': FixedP('sigmoid'),
'coef0': RealP(uniform, low=0.0, high=100),
})
kernel_nonlinear = Node(
{'gamma': RealP(uniform, low=0.00001, high=100)},
children=[kernel_poly, kernel_rbf, kernel_sigmoid])
top = Node(
{
'C':
RealP(uniform, low=1e-4, high=100),
'shrinking':
CatP(choice, items=[True, False]),
'probability':
FixedP(False),
'tol':
OrdP(choice,
items=[
0.000001, 0.00001, 0.0001, 0.001, 0.01, 0.1, 1, 10,
100, 1000, 10000
]),
'class_weight':
CatP(choice, items=[None, 'balanced']),
# 'verbose': [False],
'max_iter':
FixedP(1000000),
'decision_function_shape':
CatP(choice, items=['ovr', 'ovo'])
},
children=[kernel_linear, kernel_nonlinear])
return TransformerCS(nodes=[top])
def __new__(cls,
*args,
fields=None,
engine="dump",
db='/tmp/cururu',
settings=None,
blocking=False,
seed=0,
transformers=None):
"""Shortcut to create a ConfigSpace."""
if transformers is None:
transformers = args
if all([isinstance(t, Transformer) for t in transformers]):
return object.__new__(cls)
node = Node(
params={
'fields': FixedP(fields),
'engine': FixedP(engine),
'db': FixedP(db),
'settings': FixedP(settings),
'blocking': FixedP(blocking),
'seed': FixedP(seed),
})
return ContainerCS(Cache.name, Cache.path, transformers, nodes=[node])
def _cs_impl(cls):
params = {
'engine': CatP(choice, items=['dump', 'mysql', 'sqlite']),
'settings': FixedP({})
}
return TransformerCS(Node(params=params))
def _cs_impl(cls):
params = {'path': FixedP('./'), 'name': FixedP('iris.arff')}
return TransformerCS(Node(params=params))
def _cs_impl(cls):
params = {"path": FixedP("./"), "name": FixedP("iris.arff")}
return CS(nodes=[Node(params)])
def _cs_impl(cls):
params = {'name': FixedP('iris_OFÆdñO')}
return TransformerCS(nodes=[Node(params=params)])