-
Notifications
You must be signed in to change notification settings - Fork 0
/
bench_mark.py
168 lines (150 loc) · 9.05 KB
/
bench_mark.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
import lightgbm as lgb
import numpy as np
import pandas as pd
import utils
import tqdm
from copy import deepcopy
"""
Base tuned configs from https://github.com/catboost/catboost/tree/master/catboost/benchmarks/quality_benchmarks/notebooks
"""
BASE_CONFIGS = {
"adult": {'num_leaves': 16, 'verbose': -1, 'bagging_seed': 0, 'metric': 'auc', 'data_random_seed': 0,
'min_data_in_leaf': 2, 'bagging_fraction': 0.8231258721410332,
'min_sum_hessian_in_leaf': 0.0023527955721999202, 'feature_fraction_seed': 0,
'lambda_l1': 0.0009040060631182545, 'bagging_freq': 1, 'lambda_l2': 0, 'objective': 'binary',
'drop_seed': 0, 'learning_rate': 0.01623096612907187, 'feature_fraction': 0.6153146698674108},
'amazon': {'num_leaves': 35, 'verbose': -1, 'bagging_seed': 0, 'metric': 'auc', 'data_random_seed': 0,
'min_data_in_leaf': 43, 'bagging_fraction': 0.7533430554699614,
'min_sum_hessian_in_leaf': 5, 'feature_fraction_seed': 0, 'lambda_l1': 0,
'bagging_freq': 1, 'lambda_l2': 0, 'objective': 'binary', 'drop_seed': 0,
'learning_rate': 10 * 0.0023062503242047053, 'feature_fraction': 0.8942720034091154}, # changed learning rate to speed up convergence
'appet': {# 'num_leaves': 6, 'verbose': -1, 'bagging_seed': 0, 'metric': 'binary_logloss', 'data_random_seed': 0,
# 'min_data_in_leaf': 95, 'bagging_fraction': 0.8956806282244631,
# 'min_sum_hessian_in_leaf': 0.05112201732261002, 'feature_fraction_seed': 0,
# 'lambda_l1': 0.013807846588423104, 'bagging_freq': 1, 'lambda_l2': 0, 'objective': 'binary',
# 'drop_seed': 0, 'learning_rate': 0.012538826956019477, 'feature_fraction': 0.5112693049121645},
'metric': 'auc', 'objective': 'binary', 'bagging_freq': 1, 'feature_fraction': 0.5, 'bagging_fraction' : 0.99, 'learning_rate' : 0.05, "verbose": -1
},
'click': {'num_leaves': 45, 'verbose': -1, 'bagging_seed': 0, 'metric': 'auc', 'data_random_seed': 0,
'min_data_in_leaf': 8, 'bagging_fraction': 0.8420658067233723,
'min_sum_hessian_in_leaf': 0.004549958246639956, 'feature_fraction_seed': 0,
'lambda_l1': 0.7677413533309505, 'bagging_freq': 1, 'lambda_l2': 0.10721553735236211,
'objective': 'binary', 'drop_seed': 0, 'learning_rate': 2 * 0.029263299622061, #increased learing rate
'feature_fraction': 0.670232967018771},
'internet': {'num_leaves': 8, 'verbose': -1, 'bagging_seed': 0, 'metric': 'auc', 'data_random_seed': 0,
'min_data_in_leaf': 16, 'bagging_fraction': 0.9046010662322765,
'min_sum_hessian_in_leaf': 1.271575758364032e-05, 'feature_fraction_seed': 0,
'lambda_l1': 0.016012980104284796, 'bagging_freq': 1, 'lambda_l2': 0, 'objective': 'binary',
'drop_seed': 0, 'learning_rate': 0.006879696425995826, 'feature_fraction': 0.9957686774658463},
'kdd98' : {'num_leaves': 5, 'verbose': -1, 'bagging_seed': 0, 'metric': 'auc', 'data_random_seed': 0,
'min_data_in_leaf': 5, 'bagging_fraction': 0.7801172267397591,
'min_sum_hessian_in_leaf': 132.9945857111621, 'feature_fraction_seed': 0,
'lambda_l1': 0.0022903323397730152, 'bagging_freq': 1, 'lambda_l2': 0, 'objective': 'binary',
'drop_seed': 0, 'learning_rate': 0.029609632447460447, 'feature_fraction': 0.7235330841303137},
'kddchurn': {'num_leaves': 19, 'verbose': -1, 'bagging_seed': 0, 'metric': 'auc', 'data_random_seed': 0,
'min_data_in_leaf': 2, 'bagging_fraction': 0.7790653522214104,
'min_sum_hessian_in_leaf': 1.7132439204757702e-06, 'feature_fraction_seed': 0,
'lambda_l1': 4.31311677117027e-06, 'bagging_freq': 1, 'lambda_l2': 3.289808221335635,
'objective': 'binary', 'drop_seed': 0, 'learning_rate': 0.002468127789150438,
'feature_fraction': 0.655578595572939},
"kick" : {'num_leaves': 157, 'verbose': -1, 'bagging_seed': 0, 'metric': 'auc', 'data_random_seed': 0,
'min_data_in_leaf': 4, 'bagging_fraction': 0.7618075391294016,
'min_sum_hessian_in_leaf': 3.776016815653153e-06, 'feature_fraction_seed': 0, 'lambda_l1': 0,
'bagging_freq': 1, 'lambda_l2': 0.07371594809127686, 'objective': 'binary', 'drop_seed': 0,
'learning_rate': 0.019483006590789694, 'feature_fraction': 0.7515374545303946}, #increased learing rate
"upsel" : {'num_leaves': 11, 'verbose': -1, 'bagging_seed': 0, 'metric': 'auc', 'data_random_seed': 0,
'min_data_in_leaf': 109, 'bagging_fraction': 0.8965102673774312,
'min_sum_hessian_in_leaf': 1.0201871963154937e-05, 'feature_fraction_seed': 0,
'lambda_l1': 0.006143582721183149, 'bagging_freq': 1, 'lambda_l2': 0, 'objective': 'binary',
'drop_seed': 0, 'learning_rate': 0.004440192867499371, 'feature_fraction': 0.5804298705452275}
}
class BaseExperiment(object):
def __init__(self, dataset_name):
self.dataset_name = dataset_name
def run(self):
history = []
for params, dataset, cat_names in self.generate_configs():
log = lgb.cv(params, dataset, num_boost_round=5000, early_stopping_rounds=250)
history.append({
'params': params,
'log' : log
})
return history
def generate_configs(self):
raise NotImplementedError()
class BaselineExperiment(BaseExperiment):
def __init__(self, dataset_name):
super().__init__(dataset_name)
def generate_configs(self):
df, cat_names = utils.read_dataset_by_name(self.dataset_name)
dataset = lgb.Dataset(df.drop(['Target'], axis=1), label=df['Target'])
base_config = deepcopy(BASE_CONFIGS[self.dataset_name])
base_config['force_row_wise'] = True
yield base_config, dataset, cat_names
return
class SGBExperiment(BaseExperiment):
def __init__(self, dataset_name, bagging_fraction_space = np.linspace(0.05, 0.95, 9)):
super().__init__(dataset_name)
self.bagging_fraction_space = bagging_fraction_space
def generate_configs(self):
df, cat_names = utils.read_dataset_by_name(self.dataset_name)
dataset = lgb.Dataset(df.drop(['Target'], axis=1), label=df['Target'],)
for f in tqdm.tqdm(self.bagging_fraction_space):
base_config = deepcopy(BASE_CONFIGS[self.dataset_name])
base_config['bagging_fraction'] = f
base_config['force_row_wise'] = True
yield base_config, dataset, cat_names
return
class MVSExperiment(SGBExperiment):
DATASET_LAMBDA = {
'adult': 1e1,
'amazon': 1e-5,
'click': 1.,
'internet': 1e-2,
'kick': 10.
}
def __init__(self, dataset_name, bagging_fraction_space = np.linspace(0.05, 0.95, 9),
mvs_adaptive=True, mvs_lambda=1e-2):
super().__init__(dataset_name, bagging_fraction_space)
self.mvs_adptive = mvs_adaptive
self.mvs_lambda = self.DATASET_LAMBDA[dataset_name]
def generate_configs(self):
df, cat_names = utils.read_dataset_by_name(self.dataset_name)
dataset = lgb.Dataset(df.drop(['Target'], axis=1), label=df['Target'])
for f in tqdm.tqdm(self.bagging_fraction_space):
base_config = deepcopy(BASE_CONFIGS[self.dataset_name])
base_config['boosting'] = 'mvs'
base_config['bagging_fraction'] = f
base_config["mvs_max_sequential_size"] = 25000
base_config['mvs_adaptive'] = self.mvs_adptive
base_config['mvs_lambda'] = self.mvs_lambda
base_config['force_row_wise'] = True
yield base_config, dataset, cat_names
return
class GOSSExperiment(BaseExperiment):
DATASET_RAITO = {
'adult': (1, 5),
'amazon': (5, 1),
'click': (1, 5),
'internet': (1, 5),
'kick': (1, 5)
}
def __init__(self, dataset_name, bagging_fraction_space = np.linspace(0.05, 0.95, 9), top=1, other=5):
super().__init__(dataset_name)
self.bagging_fraction_space = bagging_fraction_space
self.top = self.DATASET_RAITO[dataset_name][0]
self.other = self.DATASET_RAITO[dataset_name][1]
def generate_configs(self):
df, cat_names = utils.read_dataset_by_name(self.dataset_name)
dataset = lgb.Dataset(df.drop(['Target'], axis=1), label=df['Target'])
for f in tqdm.tqdm(self.bagging_fraction_space):
base_config = deepcopy(BASE_CONFIGS[self.dataset_name])
base_config['boosting'] = 'goss'
# base_config['force_row_wise'] = True
base_config['top_rate'] = f * self.top / float(self.top + self.other)
base_config['other_rate'] = f * self.other / float(self.top + self.other)
base_config['bagging_freq'] = 0
base_config['bagging_fraction'] = 1.0
base_config['force_row_wise'] = True
yield base_config, dataset, cat_names
return