def init_hyperparameters(self, trial, X, y):
self.name = id_name('MultinomialNB_')
#self.classes_ = np.unique(y.astype(int))
self.alpha = trial.suggest_loguniform(self.name + "alpha", 1e-2, 100)
self.fit_prior = trial.suggest_categorical(self.name + "fit_prior",
[True, False])
def generate_hyperparameters(self, space_gen, depending_node=None):
self.name = id_name('HistGradientBoostingClassifier_')
space_gen.generate_number(self.name + "learning_rate",
0.1,
depending_node=depending_node)
space_gen.generate_number(self.name + "min_samples_leaf",
20,
depending_node=depending_node)
space_gen.generate_number(self.name + "max_leaf_nodes",
31,
depending_node=depending_node)
space_gen.generate_number(self.name + "l2_regularization",
1E-10,
depending_node=depending_node)
space_gen.generate_cat(self.name + "early_stop",
["off", "train", "valid"],
"off",
depending_node=depending_node)
space_gen.generate_number(self.name + "n_iter_no_change",
10,
depending_node=depending_node)
space_gen.generate_number(self.name + "validation_fraction",
0.1,
depending_node=depending_node)
space_gen.generate_number(self.name + "max_iter",
100,
depending_node=depending_node)
def init_hyperparameters(self, trial, X, y):
self.name = id_name('GaussianRandomProjection_')
self.n_components_fraction = trial.suggest_uniform(
self.name + 'n_components_fraction', 0.0, 1.0)
self.sparse = False
def init_hyperparameters(self, trial, X, y):
self.name = id_name('OneHotEncoder_')
self.handle_unknown = 'ignore'
self.sparse = trial.suggest_categorical(self.name + 'sparse',
[True, False])
def init_hyperparameters(self, trial, X, y):
self.name = id_name('HistGradientBoostingClassifier_')
self.loss = "auto"
self.learning_rate = trial.suggest_loguniform(
self.name + "learning_rate", 0.01, 1)
self.min_samples_leaf = trial.suggest_int(self.name +
"min_samples_leaf",
1,
200,
log=True)
self.max_depth = None
self.max_leaf_nodes = trial.suggest_int(self.name + "max_leaf_nodes",
3,
2047,
log=True)
self.max_bins = 255
self.l2_regularization = trial.suggest_loguniform(
self.name + "l2_regularization", 1E-10, 1)
self.early_stop = trial.suggest_categorical(self.name + "early_stop",
["off", "train", "valid"])
self.tol = 1e-7
self.scoring = "loss"
self.vn_iter_no_change = trial.suggest_int(
self.name + "n_iter_no_change", 1, 20)
self.validation_fraction = trial.suggest_uniform(
self.name + "validation_fraction", 0.01, 0.4)
self.classes_ = np.unique(y.astype(int))
#new
self.max_iter = trial.suggest_int(self.name + "max_iter",
10,
512,
log=True)
def generate_hyperparameters(self, space_gen, depending_node=None):
self.name = id_name('SVC_')
space_gen.generate_number(self.name + "C",
1.0,
depending_node=depending_node)
category_kernel = space_gen.generate_cat(self.name + "kernel",
["rbf", "poly", "sigmoid"],
"rbf",
depending_node=depending_node)
space_gen.generate_number(self.name + "degree",
3,
depending_node=category_kernel[1])
space_gen.generate_number(self.name + "gamma",
0.1,
depending_node=depending_node)
space_gen.generate_number(
self.name + "coef0", 0,
depending_node=depending_node) #todo: fix if we change to graph
space_gen.generate_cat(self.name + "shrinking", [True, False],
True,
depending_node=depending_node)
space_gen.generate_number(self.name + "tol",
1e-3,
depending_node=depending_node)
def init_hyperparameters(self, trial, X, y, name_space=None):
self.name = id_name('RandomForestClassifier_')
if type(name_space) != type(None):
self.name += '_' + name_space
self.criterion = trial.suggest_categorical(self.name + "criterion",
["gini", "entropy"])
self.max_features = trial.suggest_uniform(self.name + "max_features",
0., 1.)
self.max_depth = None
self.min_samples_split = trial.suggest_int(self.name +
"min_samples_split",
2,
20,
log=False)
self.min_samples_leaf = trial.suggest_int(self.name +
"min_samples_leaf",
1,
20,
log=False)
self.min_weight_fraction_leaf = 0.
self.max_leaf_nodes = None
self.min_impurity_decrease = 0.0
self.bootstrap = trial.suggest_categorical(self.name + "bootstrap",
[True, False])
#self.classes_ = 2
#hyperopt config
#self.n_estimators = trial.suggest_int(self.name + "n_estimators", 10, 512, log=True)
self.n_estimators = trial.suggest_int(self.name + "n_estimators",
10,
100,
log=True)
def init_hyperparameters(self, trial, X, y):
self.name = id_name('ExtraTreesClassifier_')
self.criterion = trial.suggest_categorical(self.name + "criterion",
["gini", "entropy"])
self.max_features = trial.suggest_uniform(self.name + "max_features",
0., 1.)
self.min_samples_split = trial.suggest_int(self.name +
"min_samples_split",
2,
20,
log=False)
self.min_samples_leaf = trial.suggest_int(self.name +
"min_samples_leaf",
1,
20,
log=False)
self.max_depth = None
self.min_weight_fraction_leaf = 0.
self.max_leaf_nodes = None
self.min_impurity_decrease = 0.0
self.bootstrap = trial.suggest_categorical(self.name + "bootstrap",
[True, False])
self.classes_ = np.unique(y.astype(int))
# hyperopt config
self.n_estimators = trial.suggest_int(self.name + "n_estimators",
10,
512,
log=True)
def generate_hyperparameters(self, space_gen, depending_node=None):
self.name = id_name('FeatureAgglomeration_')
space_gen.generate_number(self.name + 'n_components_fraction', 0.5, depending_node=depending_node)
space_gen.generate_cat(self.name + "linkage", ["ward", "complete", "average"], "ward", depending_node=depending_node)
space_gen.generate_cat(self.name + "affinity", ["euclidean", "manhattan", "cosine"], "euclidean", depending_node=depending_node)
space_gen.generate_cat(self.name + "pooling_func", [np.mean, np.median, np.max], np.mean, depending_node=depending_node)
def init_hyperparameters(self, trial, X, y):
self.name = id_name('TruncatedSVD_')
self.n_components_fraction = trial.suggest_uniform(
self.name + 'n_components_fraction', 0.0, 1.0)
self.sparse = False
def init_hyperparameters(self, trial, X, y):
self.name = id_name('PolynomialFeatures_')
self.degree = trial.suggest_int(self.name + "degree", 2, 3)
self.interaction_only = trial.suggest_categorical(self.name + "interaction_only", [False, True])
self.include_bias = trial.suggest_categorical(self.name + "include_bias", [True, False])
self.sparse = False
def generate_hyperparameters(self, space_gen, depending_node=None):
self.name = id_name('MultinomialNB_')
space_gen.generate_number(self.name + "alpha",
1,
depending_node=depending_node)
space_gen.generate_cat(self.name + "fit_prior", [True, False],
True,
depending_node=depending_node)
def generate_hyperparameters(self, space_gen, depending_node=None):
self.name = id_name('PrivateLogisticRegression_')
space_gen.generate_number(self.name + "tol",
1e-4,
depending_node=depending_node)
space_gen.generate_number(self.name + "C",
1.0,
depending_node=depending_node)
def generate_hyperparameters(self, space_gen, depending_node=None):
self.name = id_name('RBFSampler_')
space_gen.generate_number(self.name + "gamma",
1.0,
depending_node=depending_node)
space_gen.generate_number(self.name + "n_components",
100,
depending_node=depending_node)
def init_hyperparameters(self, trial, X, y):
self.name = id_name('PCA_')
self.n_components = trial.suggest_uniform(self.name + "n_components",
0.5, 0.9999)
self.whiten = trial.suggest_categorical(self.name + "whiten",
[False, True])
self.sparse = False
def generate_hyperparameters(self, space_gen, depending_node=None):
self.name = id_name('PCA_')
space_gen.generate_number(self.name + "n_components",
0.9999,
depending_node=depending_node)
space_gen.generate_cat(self.name + "whiten", [False, True],
False,
depending_node=depending_node)
def init_hyperparameters(self, trial, X, y):
self.name = id_name('RBFSampler_')
self.gamma = trial.suggest_loguniform(self.name + "gamma",
3.0517578125e-05, 8)
self.n_components = trial.suggest_int(self.name + "n_components",
50,
10000,
log=True)
self.sparse = False
def init_hyperparameters(self, trial, X, y):
self.name = id_name('PassiveAggressive_')
self.C = trial.suggest_loguniform(self.name + "C", 1e-5, 10)
self.fit_intercept = True
self.loss = trial.suggest_categorical(self.name + "loss",
["hinge", "squared_hinge"])
self.tol = trial.suggest_loguniform(self.name + "tol", 1e-5, 1e-1)
self.average = trial.suggest_categorical(self.name + "average",
[False, True])
self.max_iter = trial.suggest_int(self.name + "max_iter",
10,
1024,
log=False)
def generate_hyperparameters(self, space_gen, depending_node=None):
self.name = id_name('LinearSVC_')
space_gen.generate_cat(self.name + "penalty", ["l1", "l2"],
"l2",
depending_node=depending_node)
space_gen.generate_cat(self.name + "loss", ["hinge", "squared_hinge"],
"squared_hinge",
depending_node=depending_node)
space_gen.generate_number(self.name + "tol",
1e-4,
depending_node=depending_node)
space_gen.generate_number(self.name + "C",
1.0,
depending_node=depending_node)
def generate_hyperparameters(self, space_gen, depending_node=None):
self.name = id_name('DecisionTreeClassifier_')
space_gen.generate_cat(self.name + "criterion", ["gini", "entropy"],
"gini",
depending_node=depending_node)
space_gen.generate_number(self.name + 'max_depth_factor',
0.5,
depending_node=depending_node)
space_gen.generate_number(self.name + "min_samples_split",
2,
depending_node=depending_node)
space_gen.generate_number(self.name + "min_samples_leaf",
1,
depending_node=depending_node)
def generate_hyperparameters(self, space_gen, depending_node=None):
self.name = id_name('AdaBoostClassifier_')
space_gen.generate_number(self.name + "n_estimators",
50,
depending_node=depending_node)
space_gen.generate_number(self.name + "learning_rate",
0.1,
depending_node=depending_node)
space_gen.generate_cat(self.name + "algorithm", ["SAMME.R", "SAMME"],
"SAMME.R",
depending_node=depending_node)
space_gen.generate_number(self.name + "max_depth",
1,
depending_node=depending_node)
def init_hyperparameters(self, trial, X, y):
self.name = id_name('FeatureAgglomeration_')
self.n_components_fraction = trial.suggest_uniform(self.name + 'n_components_fraction', 0.0, 1.0)
self.linkage = trial.suggest_categorical(self.name + "linkage", ["ward", "complete", "average"])
if self.linkage == 'ward':
self.affinity = "euclidean"
else:
self.affinity = trial.suggest_categorical(self.name + "affinity", ["euclidean", "manhattan", "cosine"])
self.pooling_func = trial.suggest_categorical(self.name + "pooling_func", [np.mean, np.median, np.max])
self.sparse = False
def init_hyperparameters(self, trial, X, y):
self.name = id_name('AdaBoostClassifier_')
self.n_estimators = trial.suggest_int(self.name + "n_estimators",
50,
500,
log=False)
self.learning_rate = trial.suggest_loguniform(
self.name + "learning_rate", 0.01, 2)
self.algorithm = trial.suggest_categorical(self.name + "algorithm",
["SAMME.R", "SAMME"])
self.max_depth = trial.suggest_int(self.name + "max_depth",
1,
10,
log=False)
self.base_estimator = DecisionTreeClassifier(max_depth=self.max_depth)
self.classes_ = np.unique(y.astype(int))
def generate_hyperparameters(self, space_gen, depending_node=None):
self.name = id_name('SGDClassifier_')
space_gen.generate_cat(self.name + "loss", ["hinge", "log", "modified_huber", "squared_hinge", "perceptron"], "log", depending_node=depending_node)
space_gen.generate_cat(self.name + "penalty", ["l1", "l2", "elasticnet"], "l2", depending_node=depending_node)
space_gen.generate_number(self.name + "alpha", 0.0001, depending_node=depending_node)
space_gen.generate_number(self.name + "l1_ratio", 0.15, depending_node=depending_node)
space_gen.generate_number(self.name + "tol", 1e-4, depending_node=depending_node)
space_gen.generate_number(self.name + "epsilon", 1e-4, depending_node=depending_node)
category_learn = space_gen.generate_cat(self.name + "learning_rate", ["optimal", "invscaling", "constant"], "invscaling", depending_node=depending_node)
space_gen.generate_number(self.name + "eta0", 0.01, depending_node=depending_node) #fix if we change to graph
space_gen.generate_number(self.name + "power_t", 0.5, depending_node=category_learn[1])
space_gen.generate_cat(self.name + "average", [False, True], False, depending_node=depending_node)
space_gen.generate_number(self.name + "max_iter", 1000, depending_node=depending_node)
def init_hyperparameters(self, trial, X, y):
self.name = id_name('SparseRandomProjection_')
self.n_components_fraction = trial.suggest_uniform(
self.name + 'n_components_fraction', 0.0, 1.0)
if trial.suggest_categorical(self.name + 'density_auto',
[True, False]):
self.density = 'auto'
else:
self.density = trial.suggest_uniform(self.name + 'density_def',
1e-6, 1)
self.dense_output = trial.suggest_categorical(
self.name + 'dense_output', [True, False])
self.sparse = not self.dense_output
def init_hyperparameters(self, trial, X, y):
self.name = id_name('FastICA_')
self.algorithm = trial.suggest_categorical(self.name + 'algorithm',
['parallel', 'deflation'])
self.whiten = trial.suggest_categorical(self.name + 'whiten',
[False, True])
self.n_components_fraction = None
if self.whiten == True:
self.n_components_fraction = trial.suggest_uniform(
self.name + 'n_components_fraction', 0.0, 1.0)
self.fun = trial.suggest_categorical(self.name + 'fun',
['logcosh', 'exp', 'cube'])
self.sparse = False
def init_hyperparameters(self, trial, X, y):
self.name = id_name('RandomTreesEmbedding_')
self.n_estimators = trial.suggest_int(self.name + "n_estimators", 10,
100)
self.max_depth = trial.suggest_int(self.name + "max_depth", 2, 10)
self.min_samples_split = trial.suggest_int(
self.name + "min_samples_split", 2, 20)
self.min_samples_leaf = trial.suggest_int(
self.name + "min_samples_leaf", 1, 20)
self.min_weight_fraction_leaf = 0.0
self.max_leaf_nodes = None
self.bootstrap = trial.suggest_categorical(self.name + 'bootstrap',
[True, False])
self.sparse_output = trial.suggest_categorical(
self.name + 'sparse_output', [True, False])
self.sparse = self.sparse_output
def init_hyperparameters(self, trial, X, y):
self.name = id_name('ConstructionTransformer_')
#t = ConstructionTransformer(c_max=3, epsilon=-np.inf, scoring=auc, n_jobs=4, model=LogisticRegression(), cv=2, feature_names=attribute_names, feature_is_categorical=categorical_indicator,parameter_grid={'penalty': ['l2'], 'C': [1], 'solver': ['lbfgs'], 'class_weight': ['balanced'], 'max_iter': [10], 'multi_class':['auto']})
self.c_max = trial.suggest_int(self.name + "c_max", 2, 4)
self.epsilon = -np.inf
self.cv = 2
self.model = LogisticRegression()
self.parameter_grid = {
'penalty': ['l2'],
'C': [1],
'solver': ['lbfgs'],
'class_weight': ['balanced'],
'max_iter': [10],
'multi_class': ['auto']
}
self.n_jobs = 1
def generate_hyperparameters(self, space_gen, depending_node=None):
self.name = id_name('SelectKBest')
space_gen.generate_number(self.name + "k_fraction",
0.5,
depending_node=depending_node)
category_fs = space_gen.generate_cat(self.name + 'score_func', [
'chi2', 'f_classif', 'mutual_info', 'ExtraTreesClassifier',
'LinearSVC', 'variance'
],
"chi2",
depending_node=depending_node)
tree_catgory = category_fs[3]
lr_catgory = category_fs[4]
new_name = self.name + '_' + 'ExtraTreesClassifier' + '_'
space_gen.generate_cat(new_name + "criterion", ["gini", "entropy"],
"gini",
depending_node=tree_catgory)
space_gen.generate_number(new_name + "max_features",
0.5,
depending_node=tree_catgory)
space_gen.generate_number(new_name + "min_samples_split",
2,
depending_node=tree_catgory)
space_gen.generate_number(new_name + "min_samples_leaf",
1,
depending_node=tree_catgory)
space_gen.generate_cat(new_name + "bootstrap", [True, False],
False,
depending_node=tree_catgory)
new_name = self.name + '_' + 'LinearSVC' + '_'
space_gen.generate_cat(new_name + "loss", ["hinge", "squared_hinge"],
"squared_hinge",
depending_node=lr_catgory)
space_gen.generate_number(new_name + "tol",
1e-4,
depending_node=lr_catgory)
space_gen.generate_number(new_name + "C",
1.0,
depending_node=lr_catgory)
def generate_hyperparameters(self, space_gen, depending_node=None):
self.name = id_name('SparseRandomProjection_')
space_gen.generate_number(self.name + 'n_components_fraction',
0.1,
depending_node=depending_node)
category_density = space_gen.generate_cat(
self.name + 'density_auto', [True, False],
True,
depending_node=depending_node)
space_gen.generate_number(self.name + 'density_def',
'auto',
depending_node=category_density[1])
space_gen.generate_cat(self.name + 'dense_output', [True, False],
False,
depending_node=depending_node)