-
Notifications
You must be signed in to change notification settings - Fork 14
/
05_nn.py
235 lines (195 loc) · 7.85 KB
/
05_nn.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
229
230
231
232
233
234
235
from keras.layers import *
import jieba
import multiprocessing
import pandas as pd
from gensim.models import Word2Vec
import numpy as np
import keras.backend as K
from keras.callbacks import Callback, ModelCheckpoint
from keras.models import Model
from keras.utils.np_utils import to_categorical
from keras.preprocessing.text import Tokenizer
from keras.preprocessing.sequence import pad_sequences
from sklearn.model_selection import StratifiedKFold
from sklearn.metrics import *
import ipykernel
import tensorflow as tf
from sklearn.metrics import roc_auc_score
def to_text(row):
text = []
for fea in feas:
text.append(fea + '_' + str(row[fea]))
return " ".join(text)
def train_w2v(text_list=None, output_vector='data/w2v.txt'):
"""
训练word2vec
:param text_list:文本列表
:param output_vector:词向量输出路径
:return:
"""
print("正在训练词向量。。。")
corpus = [text.split() for text in text_list]
model = Word2Vec(corpus, size=100, window=5, min_count=1, workers=multiprocessing.cpu_count())
# 保存词向量
model.wv.save_word2vec_format(output_vector, binary=False)
# sample.csv
# test_new.csv
# train.csv
train = pd.read_csv("new_data/train.csv")
train_target = pd.read_csv('new_data/train_target.csv')
train = train.merge(train_target, on='id')
test = pd.read_csv("new_data/test.csv")
# 全量数据
train['id'] = [i for i in range(len(train))]
test['target'] = [-1 for i in range(len(test))]
df = pd.concat([train, test], sort=False)
df['certPeriod'] = df['certValidStop'] - df['certValidBegin']
no_fea = ['id', 'target', 'certValidStop', 'certValidBegin']
feas = [fea for fea in df.columns if fea not in no_fea]
# print(len(feas))
# df['token_text'] = df.apply(lambda row: to_text(row), axis=1)
# texts = df['token_text'].values.tolist()
# train_w2v(texts)
#
# # 构建词汇表
# tokenizer = Tokenizer(filters='|')
# tokenizer.fit_on_texts(texts)
# word_index = tokenizer.word_index
# print("词语数量个数:{}".format(len(word_index)))
#
# # 数据
# EMBEDDING_DIM = 100
# MAX_SEQUENCE_LENGTH = len(feas)
#
# sequences = tokenizer.texts_to_sequences(texts)
# data = pad_sequences(sequences, maxlen=MAX_SEQUENCE_LENGTH)
#
# # 类别编码
# X = data[:len(train)]
# x_test = data[len(train):]
# print(X.shape)
# print(X)
# y_train = to_categorical(train['target'].values)
y = train['target'].values
y = y.astype(np.int32)
# print(y)
X_fea = np.load(open('tmp/fea_train.npy', 'rb'))
X_fea_test = np.load(open('tmp/fea_test.npy', 'rb'))
def create_text_cnn():
#
# sequence_input = Input(shape=(MAX_SEQUENCE_LENGTH,), dtype='int32')
# embedding_layer = create_embedding(word_index, 'data/w2v.txt')
# embedding_sequences = embedding_layer(sequence_input)
# conv1 = Conv1D(128, 5, activation='relu', padding='same')(embedding_sequences)
# pool1 = MaxPool1D(3)(conv1)
# conv2 = Conv1D(128, 5, activation='relu', padding='same')(pool1)
# pool2 = MaxPool1D(3)(conv2)
# conv3 = Conv1D(128, 5, activation='relu', padding='same')(pool2)
# pool3 = MaxPool1D(3)(conv3)
# flat1 = Flatten()(pool3)
# embedding_input = embedding_layer(sequence_input)
# x_context = Bidirectional(CuDNNLSTM(128, return_sequences=True))(embedding_input)
# x = Concatenate()([embedding_input, x_context])
#
# convs = []
# for kernel_size in range(1, 5):
# conv = Conv1D(128, kernel_size, activation='relu')(x)
# convs.append(conv)
# poolings = [GlobalAveragePooling1D()(conv) for conv in convs] + [GlobalMaxPooling1D()(conv) for conv in convs]
# x = Concatenate()(poolings)
convs = []
# for kernel_size in range(1, 5):
# conv = BatchNormalization()(embedding_sequences)
# conv = Conv1D(128, kernel_size, activation='relu')(conv)
# convs.append(conv)
# poolings = [GlobalMaxPooling1D()(conv) for conv in convs]
# x_concat = Concatenate()(poolings)
fea_input = Input(shape=(98,))
fea_dense = BatchNormalization()(fea_input)
fea_dense = Reshape((98, 1, 1))(fea_dense)
con2v = Conv2D(filters=16, kernel_size=5, padding='Same',
activation='relu')(fea_dense)
con2v = Conv2D(filters=16, kernel_size=5, padding='Same',
activation='relu')(con2v)
poo2v = MaxPooling2D(pool_size=2, padding='same')(con2v)
con2v = Conv2D(filters=32, kernel_size=3, padding='Same',
activation='relu')(poo2v)
con2v = Conv2D(filters=32, kernel_size=3, padding='Same',
activation='relu')(con2v)
poo2v = MaxPooling2D(pool_size=2, strides=2, padding='same')(con2v)
flat2 = Flatten()(poo2v)
merged = Dropout(0.5)(flat2)
# merged = BatchNormalization()(merged)
dense = Dense(128, activation='relu')(merged)
pred = Dense(1, activation='sigmoid')(dense)
merged = Model(fea_input, pred)
return merged
train_pred = np.zeros((len(train, ), 1))
test_pred = np.zeros((len(test), 1))
class roc_auc_callback(Callback):
def __init__(self, training_data, validation_data):
self.x = training_data[0]
self.y = training_data[1]
self.x_val = validation_data[0]
self.y_val = validation_data[1]
def on_train_begin(self, logs={}):
return
def on_train_end(self, logs={}):
return
def on_epoch_begin(self, epoch, logs={}):
return
def on_epoch_end(self, epoch, logs={}):
y_pred = self.model.predict(self.x, verbose=0)
roc = roc_auc_score(self.y, y_pred)
logs['roc_auc'] = roc_auc_score(self.y, y_pred)
logs['norm_gini'] = (roc_auc_score(self.y, y_pred) * 2) - 1
y_pred_val = self.model.predict(self.x_val, verbose=0)
roc_val = roc_auc_score(self.y_val, y_pred_val)
logs['roc_auc_val'] = roc_auc_score(self.y_val, y_pred_val)
logs['norm_gini_val'] = (roc_auc_score(self.y_val, y_pred_val) * 2) - 1
print('\rroc_auc: %s - roc_auc_val: %s - norm_gini: %s - norm_gini_val: %s' % (
str(round(roc, 5)), str(round(roc_val, 5)), str(round((roc * 2 - 1), 5)), str(round((roc_val * 2 - 1), 5))),
end=10 * ' ' + '\n')
return
def on_batch_begin(self, batch, logs={}):
return
def on_batch_end(self, batch, logs={}):
return
skf = StratifiedKFold(n_splits=5, random_state=52, shuffle=True)
for i, (train_index, valid_index) in enumerate(skf.split(X_fea, y)):
print("n@:{}fold".format(i + 1))
# X_train = X[train_index]
# X_valid = X[valid_index]
#
X_fea_train = X_fea[train_index]
X_fea_valid = X_fea[valid_index]
y_train = y[train_index]
y_valid = y[valid_index]
model = create_text_cnn()
model.compile(loss='binary_crossentropy',
optimizer='rmsprop',
metrics=['acc'])
model.summary()
checkpoint = ModelCheckpoint(filepath='models/cnn_text_{}.h5'.format(i + 1),
monitor='val_loss',
verbose=1, save_best_only=True)
history = model.fit(X_fea_train, y_train,
validation_data=(X_fea_valid, y_valid),
epochs=10, batch_size=64,
callbacks=[checkpoint, roc_auc_callback(training_data=(X_fea_train, y_train),
validation_data=(X_fea_valid, y_valid))])
# model.load_weights('models/cnn_text.h5')
train_pred[valid_index, :] = model.predict( X_fea_valid)
test_pred += model.predict( X_fea_test)
test['target'] = test_pred / 5
test[['id', 'target']].to_csv('result/fea_cnn.csv', index=None)
# 训练数据预测结果
# 概率
# oof_df = pd.DataFrame(train_pred)
# train = pd.concat([train, oof_df], axis=1)
# # 标签
# targets = np.argmax(train_pred, axis=1)
train['pred'] = train_pred
# 分类报告
train[['id', 'target', 'pred']].to_excel('result/train.xlsx', index=None)
print(roc_auc_score(train['target'].values, train['pred'].values))