def __init__(self, info):
super(APPNPNet, self).__init__()
self.info = info
self.best_score = 0
self.hist_score = []
self.best_preds = None
self.current_round_best_preds = None
self.best_valid_score = 0
self.max_patience = 100
self.max_epochs = 1600
self.name = 'APPNP'
self.hyperparameters = {
'num_layers': self.info['num_layers'],
'lr': 0.005,
'K': 10,
'alpha': 0.15,
'dropedge_rate': self.info['dropedge_rate'],
'dropout_rate': self.info['dropout_rate'],
'hidden': self.info['init_hidden_size']
}
self.best_hp = None
self.tuner = HyperoptTuner(algorithm_name='tpe',
optimize_mode='maximize')
search_space = {
"dropedge_rate": {
"_type": "uniform",
"_value": [0, 1]
},
"dropout_rate": {
"_type": "uniform",
"_value": [0, 1]
},
"num_layers": {
"_type": "randint",
"_value": [2, 3]
},
"hidden": {
"_type": "quniform",
"_value": [4, 7, 1]
},
"lr": {
"_type": "choice",
"_value": [self.info['lr']]
},
'K': {
"_type": "quniform",
"_value": [1, 6, 1]
},
'alpha': {
"_type": "uniform",
"_value": [0, 1]
}
}
self.tuner.update_search_space(search_space)
def test_tpe(self):
tuner_fn = lambda: HyperoptTuner("tpe")
self.search_space_test_all(tuner_fn,
ignore_types=[
"uniform_equal", "qloguniform_equal",
"loguniform_equal", "quniform_clip_2"
])
# NOTE: types are ignored because `tpe.py line 465, in adaptive_parzen_normal assert prior_sigma > 0`
self.import_data_test(tuner_fn)
def __init__(self, info):
super(TAG, self).__init__()
self.info = info
self.hyperparameters = {
'num_layers': self.info['num_layers'],
'lr': self.info['lr'],
'dropedge_rate': self.info['dropedge_rate'],
'dropout_rate': self.info['dropout_rate'],
'K': 3,
'hidden': self.info['init_hidden_size'],
'use_linear': self.info['use_linear']
}
self.best_score = 0
self.hist_score = []
self.best_preds = None
self.current_round_best_preds = None
self.best_valid_score = 0
self.max_patience = 100
self.max_epochs = 1600
self.name = 'TAG'
self.best_hp = None
self.tuner = HyperoptTuner(algorithm_name='tpe',
optimize_mode='maximize')
search_space = {
"dropedge_rate": {
"_type": "choice",
"_value": [self.info['dropedge_rate']]
},
"dropout_rate": {
"_type": "choice",
"_value": [self.info['dropout_rate']]
},
"num_layers": {
"_type": "quniform",
"_value": [1, 3, 1]
},
"hidden": {
"_type": "quniform",
"_value": [4, 7, 1]
},
"lr": {
"_type": "choice",
"_value": [0.005]
},
'K': {
"_type": "quniform",
"_value": [1, 6, 1]
},
"use_linear": {
"_type": "choice",
"_value": [True, False]
}
}
self.tuner.update_search_space(search_space)
def test_tuner_generate(self):
for algorithm in ["tpe", "random_search", "anneal"]:
tuner = HyperoptTuner(algorithm)
choice_list = ["a", "b", 1, 2]
tuner.update_search_space({
"a": {
"_type": "randint",
"_value": [1, 3]
},
"b": {
"_type": "choice",
"_value": choice_list
}
})
for k in range(30):
# sample multiple times
param = tuner.generate_parameters(k)
print(param)
self.assertIsInstance(param["a"], int)
self.assertGreaterEqual(param["a"], 1)
self.assertLessEqual(param["a"], 2)
self.assertIn(param["b"], choice_list)
def test_anneal(self):
tuner_fn = lambda: HyperoptTuner("anneal")
self.search_space_test_all(tuner_fn)
self.import_data_test(tuner_fn)
def test_random_search(self):
tuner_fn = lambda: HyperoptTuner("random_search")
self.search_space_test_all(tuner_fn)
self.import_data_test(tuner_fn)
def test_anneal(self):
self.search_space_test_all(lambda: HyperoptTuner("anneal"))
def test_random_search(self):
self.search_space_test_all(lambda: HyperoptTuner("random_search"))
def test_tpe(self):
self.search_space_test_all(lambda: HyperoptTuner("tpe"))
def test_tpe(self):
self.search_space_test_all(lambda: HyperoptTuner("tpe"),
ignore_types=[
"uniform_equal", "qloguniform_equal",
"loguniform_equal", "quniform_clip_2"
])