-
Notifications
You must be signed in to change notification settings - Fork 0
/
optimizing.py
224 lines (187 loc) 路 9.69 KB
/
optimizing.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
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
from processing import Processing
from IPython.display import clear_output
from logger import logging
import psutil
import os
import joblib
import json
import gc
from xgboost import XGBClassifier
from time import sleep
import optuna
from sklearn.datasets import make_classification
from utils import get_models_folder_path
# Additional sklearn metric imports
from sklearn.metrics import (
roc_auc_score,
precision_recall_fscore_support,
average_precision_score
)
from imblearn.metrics import geometric_mean_score
class OptunaInstance(Processing):
def __init__(self, X_train, y_train, X_test, y_test, custom_param_file=None):
super().__init__(X_train, y_train)
self.X_train, self.y_train = X_train, y_train
self.X_test, self.y_test = X_test, y_test
self.custom_param_file = custom_param_file
if self.custom_param_file:
logging.warning("You are using custom hyperparameter values")
self.load_custom_params()
else:
self.param = self.get_default_params()
self.OPT_DIR_SAVE = get_models_folder_path("optuna")
def get_ram(self):
return psutil.virtual_memory().percent
def save_custom_params(self):
with open(self.custom_param_file, 'w') as json_file:
json.dump(self.param, json_file)
def load_custom_params(self):
with open(self.custom_param_file, 'r') as json_file:
self.param = json.load(json_file)
def save_study(self, study, frozen_trial):
study_path = os.path.join(self.OPT_DIR_SAVE, "xgb_optuna_study_callbacks.pkl")
joblib.dump(study, study_path)
def get_default_params(self):
default_param = {
'lambda': (1e-3, 10.0),
'alpha': (1e-3, 10.0),
'gamma':([0,1,5]),
'objective': (['binary:logistic']),
'min_child_weight': (1, 10),
'reg_alpha': (0, 1),
'reg_lambda': (0, 1),
"scale_pos_weight": (1, 10),
# 'max_features': ('max_features', ['auto', 'sqrt', 'log2']),
'colsample_bytree': ([0.6,0.7,0.8,0.9,1.0]),
'subsample': ([0.6,0.7,0.8,0.9,1.0]),
'learning_rate': ([0.008, 0.01, 0.03, 0.05, 0.07, 0.1]),
'max_depth': ([9,11,13]),
'random_state': ([48]),
# 'eval_metric': trial.suggest_categorical('eval_metric', [['auc','error']]),
}
return default_param
# def get_models_folder_path(self):
# current_folder = os.path.dirname(os.path.abspath(__file__))
# models_folder_path = os.path.join(current_folder, "optuna")
# if not os.path.exists(models_folder_path):
# os.makedirs(models_folder_path)
# return models_folder_path
def logging_callback(self, study, frozen_trial):
previous_best_value = study.user_attrs.get("previous_best_value", None)
if previous_best_value != study.best_value:
study.set_user_attr("previous_best_value", study.best_value)
clear_output(wait=True)
self.text_out = f"Trial {frozen_trial.number} done with best value: {frozen_trial.value} and parameters: {frozen_trial.params}."
logging.info(self.text_out)
# Writting to file
with open(f"{self.OPT_DIR_SAVE}/xgb_optuna_study_log.txt", "a") as os_log:
os_log.write('\n\n')
os_log.write(f"Current Ram Used: {self.get_ram()} %\n")
os_log.write(self.text_out)
def objective(self, trial, n_estimators=500, tree_method="hist"):
nn_early_stopping_rounds = n_estimators * 0.1
# access the hyperparamater
lambda_range = self.param.get('lambda', (1e-3, 10.0))
alpha_range = self.param.get('alpha', (1e-3, 10.0))
gamma_range = self.param.get('gamma', [0, 1, 5])
min_child_weight_range = self.param.get('min_child_weight', (1, 10))
reg_alpha_range = self.param.get('reg_alpha', (0, 1))
reg_lambda_range = self.param.get('reg_lambda', (0, 1))
scale_pos_weight_range = self.param.get('scale_pos_weight', (1, 10))
colsample_bytree_range = self.param.get('colsample_bytree', [0.6,0.7,0.8,0.9,1.0])
subsample_range = self.param.get('subsample', [0.6,0.7,0.8,0.9,1.0])
learning_rate_range = self.param.get('learning_rate', [0.008, 0.01, 0.03, 0.05, 0.07, 0.1])
max_depth_range = self.param.get('max_depth', [9,11,13])
param = {
'tree_method':trial.suggest_categorical('tree_method', [tree_method]),
'lambda': trial.suggest_float('lambda', *lambda_range),
'alpha': trial.suggest_float('alpha', *alpha_range),
'gamma':trial.suggest_categorical('gamma', gamma_range),
'objective': trial.suggest_categorical('objective', ['binary:logistic']),
'min_child_weight': trial.suggest_int('min_child_weight', *min_child_weight_range),
'reg_alpha': trial.suggest_float('reg_alpha', *reg_alpha_range),
'reg_lambda': trial.suggest_float('reg_lambda', *reg_lambda_range),
"scale_pos_weight": trial.suggest_float('scale_pos_weight', *scale_pos_weight_range),
# 'max_features': trial.suggest_categorical('max_features', ['auto', 'sqrt', 'log2']),
'colsample_bytree': trial.suggest_categorical('colsample_bytree',colsample_bytree_range),
'subsample': trial.suggest_categorical('subsample', subsample_range),
'learning_rate': trial.suggest_categorical('learning_rate',learning_rate_range),
'n_estimators': trial.suggest_categorical('n_estimators',[n_estimators]),
'max_depth': trial.suggest_categorical('max_depth', max_depth_range),
'random_state': trial.suggest_categorical('random_state', [48]),
'early_stopping_rounds': trial.suggest_categorical('early_stopping_rounds',[nn_early_stopping_rounds]),
# 'eval_metric': trial.suggest_categorical('eval_metric', [['auc','error']]),
}
if self.get_ram() >= 90:
raise MemoryError("Short On Memory")
if trial.number > 0:
clear_output(wait=True)
gc.collect()
logging.info(self.text_out)
model_xgbc = XGBClassifier(**param)
eval_setparam = [(self.X_test, self.y_test)]
logging.info(f"Current Ram Used: {self.get_ram()} %")
model_xgbc.fit(self.X_train, self.y_train, eval_set=eval_setparam, verbose=False)
preds = model_xgbc.predict(self.X_test)
roc_auc = roc_auc_score(self.y_test, preds)
g_mean = geometric_mean_score(self.y_test, preds)
precision, recall, f1_score, _ = precision_recall_fscore_support(self.y_test, preds, average='macro')
pr_auc = average_precision_score(self.y_test, preds, average='macro')
# the decision metric
max_auc_gmean=max(roc_auc, g_mean)
score = max_auc_gmean * f1_score
trial.report(score, 1)
if trial.should_prune():
text_prune = f'Trial {trial.number} pruned'
# Writing to file
with open(f"{self.OPT_DIR_SAVE}/xgb_optuna_study_log.txt", "a") as os_log:
os_log.write('\n')
os_log.write(text_prune)
del model_xgbc, preds, text_prune
gc.collect()
sleep(3)
raise optuna.TrialPruned()
text_dtl = f"Trial {trial.number} finished with parameters: {trial.params}"
# Writing to file
with open(f"{self.OPT_DIR_SAVE}/xgb_optuna_study_log.txt", "a") as os_log:
os_log.write('\n\n')
os_log.write(f"Current Ram Used: {self.get_ram()} %")
os_log.write(text_dtl)
return score
def OptunaStudy(X_train, y_train, X_test, y_test, sampling, ratio, custom_param_file=None):
ot = OptunaInstance(X_train, y_train, X_test, y_test, custom_param_file)
# ot.split_data(0.8)
# ot.sampling(sampling=sampling, ratio=ratio)
# ot.X, ot.y = None, None
# del X, y
# gc.collect()
# sleep(3)
nn_trials = 10
nn_estimators = 100
if os.path.exists(f'{ot.OPT_DIR_SAVE}/xgb_optuna_study_log_{sampling}_{ratio}.txt'):
os.remove(f'{ot.OPT_DIR_SAVE}/xgb_optuna_study_log_{sampling}_{ratio}.txt')
optuna.logging.set_verbosity(optuna.logging.WARN)
logging.info("===================HYPERPARAMETER OPTIMIZING====================")
logging.info(f"Please wait, finding best trial ...")
study = optuna.create_study(direction="maximize")
try:
# callbacks [self.save_study] is to save study in case memory fails
study.optimize(lambda trial: ot.objective(trial, n_estimators = nn_estimators),
n_trials = nn_trials,
callbacks = [ot.logging_callback, ot.save_study],
gc_after_trial = True,
catch = (RuntimeWarning,ArithmeticError,))
except MemoryError as e:
logging.info(f'{e} : Memory was getting low, Trial ended early')
complete_trials = [t for t in study.trials if t.state == optuna.trial.TrialState.COMPLETE]
pruned_trials = [t for t in study.trials if t.state == optuna.trial.TrialState.PRUNED]
logging.info(f'Number of finished trials: {len(study.trials)}')
logging.info(f'Number of pruned trials: {len(pruned_trials)}')
logging.info(f'Number of completed trials: {len(complete_trials)}')
logging.info(f'Best trial: {study.best_trial.params}')
joblib.dump(study, f"{ot.OPT_DIR_SAVE}/xgb_optuna_study_{sampling}_{ratio}.pkl")
return study
if __name__ == "__main__":
# Create an instance of the Optuna class
X, y = make_classification(n_samples=1000, n_features=5, n_classes=2, weights=[0.9, 0.1], random_state=42)
study_results = OptunaStudy(X, y, "sm", 0.5)