def get_knob_config():
return {
'max_iter': IntegerKnob(10, 40 if APP_MODE != 'DEV' else 10),
'kernel': CategoricalKnob(['rbf', 'linear']),
'gamma': CategoricalKnob(['scale', 'auto']),
'C': FloatKnob(1e-2, 1e2, is_exp=True)
}
def get_knob_config():
return {
'max_epochs': FixedKnob(10),
'learning_rate': FloatKnob(1e-5, 1e-2, is_exp=True),
'batch_size': CategoricalKnob([16, 32, 64, 128]),
'max_image_size': CategoricalKnob([32, 64, 128, 224]),
}
def get_knob_config():
return {
'max_depth': IntegerKnob(1, 32),
'splitter': CategoricalKnob(['best', 'random']),
'criterion': CategoricalKnob(['gini', 'entropy']),
'max_image_size': CategoricalKnob([16, 32])
}
def get_knob_config():
return {
'max_iter': FixedKnob(20),
'kernel': CategoricalKnob(['rbf', 'linear', 'poly']),
'gamma': CategoricalKnob(['scale', 'auto']),
'C': FloatKnob(1e-4, 1e4, is_exp=True),
'max_image_size': CategoricalKnob([16, 32])
}
def get_knob_config():
return {
'max_epochs': FixedKnob(10),
'hidden_layer_count': IntegerKnob(1, 2),
'hidden_layer_units': IntegerKnob(2, 128),
'learning_rate': FloatKnob(1e-5, 1e-1, is_exp=True),
'batch_size': CategoricalKnob([16, 32, 64, 128]),
'max_image_size': CategoricalKnob([16, 32, 48]),
'quick_train':
PolicyKnob('EARLY_STOP') # Whether early stopping would be used
}
def get_knob_config():
return {
'int': IntegerKnob(1, 32),
'float': FloatKnob(1e-5, 1),
'cat': CategoricalKnob(['a', 'b', 'c']),
'fixed': FixedKnob('fixed')
}
def test_standard_knobs(self, budget):
knob_config = {
'int': IntegerKnob(2, 128),
'float': FloatKnob(1e-5, 1e-1, is_exp=True),
'cat': CategoricalKnob([16, 32, 64, 128]),
}
advisor = make_advisor(knob_config, budget)
assert isinstance(advisor, BayesOptAdvisor)
def test_arch_knobs_with_standard_knobs(self, budget, arch_knob):
knob_config = {
'arch': arch_knob,
'float': FloatKnob(1e-5, 1e-1, is_exp=True),
'cat': CategoricalKnob([16, 32, 64, 128]),
}
advisor = make_advisor(knob_config, budget)
assert isinstance(advisor, RandomAdvisor)
def get_knob_config():
return {
'epochs': FixedKnob(10),
'word_embed_dims': IntegerKnob(16, 128),
'word_rnn_hidden_size': IntegerKnob(16, 128),
'word_dropout': FloatKnob(1e-3, 2e-1, is_exp=True),
'learning_rate': FloatKnob(1e-2, 1e-1, is_exp=True),
'batch_size': CategoricalKnob([16, 32, 64, 128]),
}
def test_standard_knobs_with_params_sharing(self, budget):
knob_config = {
'int': IntegerKnob(2, 128),
'float': FloatKnob(1e-5, 1e-1, is_exp=True),
'cat': CategoricalKnob([16, 32, 64, 128]),
'share_params': PolicyKnob('SHARE_PARAMS')
}
advisor = make_advisor(knob_config, budget)
assert isinstance(advisor, BayesOptWithParamSharingAdvisor)
def test_standard_knobs_with_early_stop(self, budget):
knob_config = {
'int': IntegerKnob(2, 128),
'float': FloatKnob(1e-5, 1e-1, is_exp=True),
'cat': CategoricalKnob([16, 32, 64, 128]),
'early_stop': PolicyKnob('EARLY_STOP')
}
advisor = make_advisor(knob_config, budget)
assert isinstance(advisor, BayesOptAdvisor)
def get_knob_config():
return {
'epochs': FixedKnob(2),
'hidden_layer_count': IntegerKnob(1, 2),
'hidden_layer_units': IntegerKnob(2, 128),
'learning_rate': FloatKnob(1e-5, 1e-1, is_exp=True),
'batch_size': CategoricalKnob([16, 32, 64, 128]),
'image_size': FixedKnob(32)
}
def get_knob_config():
return {
'trial_epochs': FixedKnob(300),
'lr': FloatKnob(1e-4, 1, is_exp=True),
'lr_decay': FloatKnob(1e-3, 1e-1, is_exp=True),
'opt_momentum': FloatKnob(0.7, 1, is_exp=True),
'opt_weight_decay': FloatKnob(1e-5, 1e-3, is_exp=True),
'batch_size': CategoricalKnob([32, 64, 128]),
'drop_rate': FloatKnob(0, 0.4),
'max_image_size': FixedKnob(32),
'share_params': PolicyKnob('SHARE_PARAMS'),
# Affects whether training is shortened by using early stopping
'quick_train': PolicyKnob('EARLY_STOP'),
'early_stop_train_val_samples': FixedKnob(1024),
'early_stop_patience_epochs': FixedKnob(5)
}
def get_knob_config():
return {
'max_depth': IntegerKnob(1, 32),
'criterion': CategoricalKnob(['gini', 'entropy'])
}
def get_knob_config():
return {
'max_depth': IntegerKnob(2, 16 if APP_MODE != 'DEV' else 4),
'criterion': CategoricalKnob(['gini', 'entropy'])
}
def get_knob_config():
return {
'epochs': IntegerKnob(1, 1 if APP_MODE != 'DEV' else 10),
'learning_rate': FloatKnob(1e-5, 1e-1, is_exp=True),
'batch_size': CategoricalKnob([16, 32, 64, 128]),
}