def __init__(self):
hp_space = [
HyperParameter.float_param('learning_rate', (1e-6, 0.15)),
HyperParameter.int_param('global_step', (0, 1e9)),
HyperParameter.float_param(
'initial_gradient_squared_accumulator_value', (1e-6, 1.)),
HyperParameter.float_param('l1_regularization_strength', (0., 1.)),
HyperParameter.float_param('l2_regularization_strength', (0., 1.)),
HyperParameter.categorical_param('use_locking', (True, False))
]
model_initializer = tf.train.AdagradDAOptimizer
super().__init__(hp_space, model_initializer)
def __init__(self):
hp_space = [
HyperParameter.float_param('nu', (5e-3, 1)),
HyperParameter.categorical_param('kernel', ('poly', 'rbf', 'sigmoid')),
HyperParameter.int_param('degree', (2, 5)),
HyperParameter.float_param('gamma', (3.0517578125e-05, 8)),
HyperParameter.float_param('coef0', (-1, 1.)),
# HyperParameter.categorical_param('shrinking', (True, False)),
HyperParameter.float_param('tol', (1e-5, 1e-1))
]
model_initializer = sklearn.svm.NuSVC
super().__init__(hp_space, model_initializer)
def __init__(self):
hp_space = [
HyperParameter.float_param('learning_rate', (1e-6, 0.15)),
HyperParameter.float_param('learning_rate_power', (-2., -1e-6)),
HyperParameter.float_param('initial_accumulator_value', (0., 2.)),
HyperParameter.float_param('l1_regularization_strength', (0., 2.)),
HyperParameter.float_param('l2_regularization_strength', (0., 2.)),
HyperParameter.categorical_param('use_locking', (True, False)),
HyperParameter.float_param('l2_shrinkage_regularization_strength',
(0., 2.)),
]
model_initializer = tf.train.FtrlOptimizer
super().__init__(hp_space, model_initializer)
def __init__(self):
hp_space = [
HyperParameter.int_param('n_estimators', (10, 1000)),
HyperParameter.int_param('max_depth', (0, 40)),
HyperParameter.int_param('min_samples_split', (1, 100)),
HyperParameter.int_param('min_samples_leaf', (1, 100)),
HyperParameter.categorical_param('max_features',
('sqrt', 'log2', None)),
HyperParameter.int_param('max_leaf_nodes', (-1, 100)),
HyperParameter.float_param('min_impurity_decrease', (0.0, 100.0))
]
initializer = sklearn.ensemble.RandomForestClassifier
super().__init__(hp_space, initializer)
def __init__(self):
hp_space = [
HyperParameter.int_param('max_depth', (0, 60)),
HyperParameter.float_param('learning_rate', (0, 1)),
HyperParameter.int_param('n_estimators', (50, 1000)),
HyperParameter.categorical_param('booster',
('gbtree', 'gblinear', 'dart')),
HyperParameter.float_param('gamma', (0, 10000)),
HyperParameter.int_param('min_child_weight', (0, 100)),
HyperParameter.int_param('max_delta_step', (0, 10)),
HyperParameter.float_param('subsample', (0.1, 1)),
HyperParameter.float_param('colsample_bytree', (0.1, 1)),
HyperParameter.float_param('colsample_bylevel', (0.1, 1)),
HyperParameter.float_param('reg_alpha', (0.0, 1e4)),
HyperParameter.float_param('reg_lambda', (0.0, 1e4)),
HyperParameter.categorical_param(
'tree_method',
('exact', 'approx', 'hist', 'gpu_exact', 'gpu_hist')),
HyperParameter.float_param('sketch_eps', (0.003, 1)),
HyperParameter.categorical_param('grow_policy',
('depthwise', 'lossguide')),
HyperParameter.int_param('max_leaves', (0, 100)),
HyperParameter.int_param('max_bin', (20, 2000)),
HyperParameter.categorical_param('sample_type',
('uniform', 'weighted')),
HyperParameter.categorical_param('normalize_type',
('tree', 'forest')),
HyperParameter.float_param('rate_drop', (0, 1)),
HyperParameter.float_param('skip_drop', (0, 1)),
HyperParameter.categorical_param('updater',
('shotgun', 'coord_descent')),
HyperParameter.categorical_param(
'feature_selector',
('cyclic', 'shuffle', 'random', 'greedy', 'thrifty'))
]
initializer = XGBClassifier
super().__init__(hp_space, initializer)
