def get_decision_tree(self, nominal_indices, **kwargs):
param_dist = {
'Estimator__min_samples_split':
loguniform_int(base=2, low=2**1, high=2**7),
'Estimator__min_samples_leaf':
loguniform_int(base=2, low=2**0, high=2**6)
}
decision_tree = self._get_pipeline(nominal_indices,
DecisionTreeClassifier())
random_search = RandomizedSearchCV(decision_tree, param_dist,
**self.rs_arguments)
return random_search
def get_uniform_paramgrid(hyperparameters, fixed_parameters):
param_grid = dict()
for param_name, hyperparameter in hyperparameters.items():
if fixed_parameters is not None and param_name in fixed_parameters.keys(
):
continue
if isinstance(hyperparameter, CategoricalHyperparameter):
all_values = hyperparameter.choices
if all(item in ['True', 'False'] for item in all_values):
all_values = [bool(item) for item in all_values]
param_grid[param_name] = all_values
elif isinstance(hyperparameter, UniformFloatHyperparameter):
if hyperparameter.log:
param_grid[param_name] = loguniform(base=2,
low=hyperparameter.lower,
high=hyperparameter.upper)
else:
param_grid[param_name] = uniform(loc=hyperparameter.lower,
scale=hyperparameter.upper -
hyperparameter.lower)
elif isinstance(hyperparameter, UniformIntegerHyperparameter):
if hyperparameter.log:
param_grid[param_name] = loguniform_int(
base=2,
low=hyperparameter.lower,
high=hyperparameter.upper)
else:
param_grid[param_name] = randint(low=hyperparameter.lower,
high=hyperparameter.upper + 1)
else:
raise ValueError()
return param_grid
def test_serialize_complex_param_grid(self):
# JvR: this unit test seems to fail..
# TODO: FIXME
parameter_grid = {
'n_estimators': [1, 5, 10, 100],
'learning_rate': loguniform(base=2, low=0.01, high=0.99),
'base_estimator__max_depth': loguniform_int(base=2, low=1, high=16)
}
serialized = openml.flows.sklearn_to_flow(parameter_grid)
deserialized = dict(openml.flows.flow_to_sklearn(serialized))
self.assertDictEqual(parameter_grid, deserialized)
def get_mlp(self, nominal_indices, **kwargs):
param_dist = {
'Estimator__hidden_layer_sizes':
loguniform_int(base=2, low=2**5, high=2**11),
'Estimator__learning_rate_init':
loguniform(base=10, low=10**-5, high=10**0),
'Estimator__alpha':
loguniform(base=10, low=10**-7, high=10**-4),
'Estimator__max_iter':
scipy.stats.randint(2, 1001),
'Estimator__momentum':
scipy.stats.uniform(loc=0.1, scale=0.8)
}
mlp = self._get_pipeline(
nominal_indices, MLPClassifier(activation='tanh', solver='adam'))
random_search = RandomizedSearchCV(mlp, param_dist,
**self.rs_arguments)
return random_search