-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathnn+gbdt.py
228 lines (186 loc) · 9.05 KB
/
nn+gbdt.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
225
226
227
228
import pandas as pd
import numpy as np
from sklearn.model_selection import KFold
from sklearn.model_selection import train_test_split
import h5py
import pickle
import sklearn
from keras.callbacks import Callback
from keras.models import Model
from sklearn.preprocessing import MinMaxScaler # 这是标准化处理的语句，很方便，里面有标准化和反标准化。。
from keras.models import Sequential
from keras.layers import Dense, Dropout, BatchNormalization
from keras.losses import binary_crossentropy
from keras.optimizers import Adam
from keras.callbacks import ModelCheckpoint, EarlyStopping
from sklearn.ensemble import GradientBoostingRegressor
import os
def smape_error(preds, train_data):
labels = train_data.get_label()
return 'error', np.mean(np.fabs(preds - labels) / (preds + labels) * 2), False
def min_max_normalize(data):
# 归一化 数据的归一化计算，这样计算之后结果能更加适合非树模型， 但是进行归一化之后怎么反归一化得看下
#数据量大，标准化慢
df=data.apply(lambda x: (x - np.min(x)) / (np.max(x) - np.min(x)))
# 做简单的平滑,试试效果如何
return df
def get_score(pred, valid_y_exp):
return np.mean(np.abs(pred - valid_y_exp) / (pred + valid_y_exp) * 2)
def NN_Model():
'''
训练nn模型，并提取，倒数第二层的特征，特征的提取方法可参照：https://blog.csdn.net/hahajinbu/article/details/77982721
进行最后一层的抽取方法是， 先训练一个nn模型model，但是要提前给每层都赋好层命名， 之后再简历一个Model,输入是上一个模型
训练所使用到的数据，输出是上一个model的指定层名，最为输出，然后使用Model去做预测，得到输出那一层结果
'''
train = pd.read_csv( './data/do_feat_data_2/train.csv')
test = pd.read_csv( './data/do_feat_data_2/test.csv')
train.pop('板温**1/2')
train.pop('现场温度**1/2')
test.pop('板温**1/2')
test.pop('现场温度**1/2')
#print('ddd:',len(test.columns))
train=train.interpolate()
train=train[~train.isin([np.nan,np.inf,-np.inf]).any(1)]
#train.replace(np.inf,999)
y = train.pop('发电量')
#test=test.interpolate()
#test=test[~test.isin([np.nan,np.inf,-np.inf]).any(1)]
# f['data'].value存放的是时间戳 上空间的流量数据
scaler = MinMaxScaler(feature_range=(-1, 1))
scaler.fit(train)
train=scaler.fit_transform(train)
#test = scaler.fit_transform(test)
scaler = MinMaxScaler(feature_range=(-1, 1))
scaler.fit(y.reshape(-1, 1))
y=scaler.fit_transform(y.reshape(-1, 1))
#print(pm25_data_all.head())
#print('000000000000000000',pm25_data_all.columns.values.tolist())
#pm25_data_nor = min_max_normalize(pm25_data_all)
#print(pm25_data_nor.columns.values.tolist())
train_x, valid_x, train_y, valid_y = train_test_split(train,y,
test_size=0.2, random_state=11)
model = Sequential()
model.add(Dense(activation='relu', units=800, input_dim=902))
model.add(BatchNormalization(axis=1))
model.add(Dense(activation='relu', units=200, name='Dense_2'))
model.add(Dense(activation='tanh', units=1))
optimizer = Adam(lr=0.00001)
model.compile(loss='mse', optimizer=optimizer)
# mc = ModelCheckpoint(filepath="./model/weights-improvement-{epoch:02d}-{val_auc:.2f}.h5", monitor='val_auc', verbose=1, save_best_only=True, save_weights_only=False, mode='max', period=0)
es = EarlyStopping(monitor='val_rmse', patience=10, verbose=1, mode='min')
model.fit(x=train_x, y=train_y, batch_size=32, epochs=200,
validation_data=(valid_x, valid_y), verbose=1, callbacks=[es])
model_file = os.getcwd() +'/model_saver/nn_1.model'
model.save_weights(model_file, overwrite=True)
# 开始进行抽取
dense1_layer_model = Model(inputs=model.input,
outputs=model.get_layer('Dense_2').output)
def use_nn_to_gbdt(train_air='pm25'):
###################### 1.书写之前的网络结构 ####################
model = Sequential()
model.add(Dense(activation='relu', units=800, input_dim=902))
model.add(BatchNormalization(axis=1))
model.add(Dense(activation='relu', units=200, name='Dense_2'))
model.add(Dense(activation='tanh', units=1))
#################### 加载网络模型，和加载全部数据，作前40W做初始化和提取 ###############
train = pd.read_csv( './data/do_feat_data_2/train.csv')
test = pd.read_csv( './data/do_feat_data_2/test.csv')
train.pop('板温**1/2')
train.pop('现场温度**1/2')
test.pop('板温**1/2')
test.pop('现场温度**1/2')
train=train.interpolate()
train=train[~train.isin([np.nan,np.inf,-np.inf]).any(1)]
test=test.interpolate()
test.replace(np.inf,999)
test.replace(-np.inf, -999)
test=test[~test.isin([np.nan,np.inf,-np.inf]).any(1)]
y = train.pop('发电量')
scaler = MinMaxScaler(feature_range=(-1, 1))
scaler.fit(train)
train=pd.DataFrame(scaler.fit_transform(train))
scaler = MinMaxScaler(feature_range=(-1, 1))
scaler.fit(y.reshape(-1, 1))
y=scaler.fit_transform(y.reshape(-1, 1))
model_file = os.getcwd() +'/model_saver/nn_1.model'
model.load_weights(model_file)
dense2_layer_model = Model(inputs=model.input,
outputs=model.get_layer('Dense_2').output)
dense2_output = dense2_layer_model.predict(train)
print(pd.DataFrame(dense2_output).head)
#拼接标签
################# 3.提取出来的特征作为GBDT的输入，重新训练一个模型 ########################
dense2_output=pd.DataFrame(dense2_output)
gbm0 = GradientBoostingRegressor()
gbm0.fit(dense2_output, y)
model_file=os.getcwd() +'/model_saver/nn_gbdt.model'
with open(model_file, 'wb') as fout:
pickle.dump(gbm0, fout)
# test_Y1 = gbm0.predict(test_X)
# score = get_score(test_Y1, test_Y)
def gbdt_nn_predict(train_air='pm25'):
###################### 1.书写之前的网络结构 ####################
model = Sequential()
model.add(Dense(activation='relu', units=800, input_dim=902))
model.add(BatchNormalization(axis=1))
model.add(Dense(activation='relu', units=200, name='Dense_2'))
model.add(Dense(activation='tanh', units=1))
###########################################################
#################### 加载网络模型，和加载全部数据，作前40W做初始化和提取 ###############
train = pd.read_csv('./data/do_feat_data_2/train.csv')
test = pd.read_csv('./data/do_feat_data_2/test.csv')
train.pop('板温**1/2')
train.pop('现场温度**1/2')
test.pop('板温**1/2')
test.pop('现场温度**1/2')
test.replace(np.inf,np.nan,inplace=True)
test.replace({-np.inf:np.nan},inplace=True)
test.replace({np.nan: 1}, inplace=True)
#使用前向填充方法
test.fillna(method='pad')
train = train.interpolate()
train = train[~train.isin([np.nan, np.inf, -np.inf]).any(1)]
test = test.interpolate()
#test.to_csv(os.getcwd() +'/model_saver/show.csv')
#print(':::::::::::',np.isinf(test))
if np.isinf(test['Unnamed: 0/风速'][8]):
print(':::::', test['Unnamed: 0*电压C'][1])
test.to_csv(os.getcwd() +'/model_saver/show.csv')
#test.replace(np.nan, 0)
#test = test[~test.isin([np.nan, np.inf, -np.inf]).any(1)]
y = train.pop('发电量')
scaler = MinMaxScaler(feature_range=(-1, 1))
scaler.fit(train)
test = pd.DataFrame(scaler.fit_transform(test))
scaler1 = MinMaxScaler(feature_range=(-1, 1))
scaler1.fit(y.reshape(-1, 1))
y = scaler1.fit_transform(y.reshape(-1, 1))
########################### 用神经网络作提取 #########################
model_file = os.getcwd() +'/model_saver/nn_1.model'
model.load_weights(model_file)
dense2_layer_model = Model(inputs=model.input,
outputs=model.get_layer('Dense_2').output)
dense2_output = dense2_layer_model.predict(test)
dense2_output = pd.DataFrame(dense2_output)
print('dense2_output:',dense2_output)
model_path=os.getcwd() +'/model_saver/nn_gbdt.model'
model = pickle.load(open(model_path, 'rb'))
yy=model.predict(dense2_output)
yy = scaler1.inverse_transform(yy.reshape(-1, 1))
print(',,,,,:',yy)
lgbm_submission = pd.read_csv('./data/base_data/submit_example.csv',header=None)
lgbm_submission.columns=['A','B']
res = pd.DataFrame()
res['A'] = list(lgbm_submission['A'])
res['B'] = list(yy)
#print(res[res['RST'] == 1].shape)
#################################### 提交和线下结果展示 #####################################
res.to_csv(os.getcwd() +'/submit/%s_%s.csv'% (str(12), str(21)),
index=False, header=None)
if __name__=='__main__':
print('************ 1.训练nn模型 *************')
NN_Model()
print('************ 2.训练gbdt模型 *************')
use_nn_to_gbdt()
print('************ 3.对test集进行预测 *************')
gbdt_nn_predict()