-
Notifications
You must be signed in to change notification settings - Fork 26
/
Copy pathwide_and_deep.py
187 lines (164 loc) · 8.08 KB
/
wide_and_deep.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
import pandas as pd
import numpy as np
from sklearn.preprocessing import StandardScaler, PolynomialFeatures, LabelEncoder
from keras.layers import Input, Embedding, Dense, Flatten, Dropout, SpatialDropout1D, Activation, concatenate
from keras.optimizers import Adam, SGD
from keras.layers.advanced_activations import ReLU, PReLU, LeakyReLU, ELU
from keras.layers.normalization import BatchNormalization
from keras.callbacks import EarlyStopping, ModelCheckpoint
from keras.models import Model
from tensorflow.keras.utils import plot_model
COLUMNS = [
"age", "workclass", "fnlwgt", "education", "education_num", "marital_status",
"occupation", "relationship", "race", "gender", "capital_gain", "capital_loss",
"hours_per_week", "native_country", "income_bracket"
]
LABEL_COLUMN = "label"
CATEGORICAL_COLUMNS = [
"workclass", "education", "marital_status", "occupation", "relationship",
"race", "gender", "native_country"
]
CONTINUOUS_COLUMNS = [
"age", "education_num", "capital_gain", "capital_loss", "hours_per_week"
]
def preprocessing():
train_data = pd.read_csv('./adult.data', names=COLUMNS)
train_data.dropna(how='any', axis=0)
test_data = pd.read_csv('./adult.test', skiprows=1, names=COLUMNS)
test_data.dropna(how='any', axis=0)
all_data = pd.concat([train_data, test_data])
# ラベルを数値化する
all_data[LABEL_COLUMN] = all_data['income_bracket'].apply(lambda x: ">50K" in x).astype(int)
all_data.pop('income_bracket')
y = all_data[LABEL_COLUMN].values
all_data.pop(LABEL_COLUMN)
for c in CATEGORICAL_COLUMNS:
le = LabelEncoder()
all_data[c] = le.fit_transform(all_data[c])
train_size = len(train_data)
x_train = all_data.iloc[:train_size]
y_train = y[:train_size]
x_test = all_data.iloc[train_size:]
y_test = y[train_size:]
x_train_categ = np.array(x_train[CATEGORICAL_COLUMNS]) # カテゴリーデータ
x_test_categ = np.array(x_test[CATEGORICAL_COLUMNS])
x_train_conti = np.array(x_train[CONTINUOUS_COLUMNS], dtype='float64') # 連続的データ
x_test_conti = np.array(x_test[CONTINUOUS_COLUMNS], dtype='float64')
scaler = StandardScaler()
x_train_conti = scaler.fit_transform(x_train_conti) # 連続データの訓練セットの平均とstdで標準化
x_test_conti = scaler.transform(x_test_conti)
return [x_train, y_train, x_test, y_test, x_train_categ, x_test_categ, x_train_conti, x_test_conti, all_data]
class Wide_and_Deep:
def __init__(self, mode='wide and deep'):
self.mode = mode
x_train, y_train, x_test, y_test, x_train_categ, x_test_categ, x_train_conti, x_test_conti, all_data \
= preprocessing()
self.x_train = x_train
self.y_train = y_train
self.x_test = x_test
self.y_test = y_test
self.x_train_categ = x_train_categ # 訓練セットの中のカテゴリーデータ
self.x_test_categ = x_test_categ # テストセットの中のカテゴリーデータ
self.x_train_conti = x_train_conti # 訓練セットの中の連続的データ
self.x_test_conti = x_test_conti # テストセットの中の連続的データ
self.all_data = all_data
self.poly = PolynomialFeatures(degree=2, interaction_only=True)
# カテゴリーデータをcross product化
self.x_train_categ_poly = self.poly.fit_transform(x_train_categ)
self.x_test_categ_poly = self.poly.transform(x_test_categ)
self.categ_inputs = None
self.conti_input = None
self.deep_component_outlayer = None
self.logistic_input = None
self.model = None
def deep_component(self):
categ_inputs = []
categ_embeds = []
# カテゴリーデータの特徴ごとにInput層とEmbedding層を作成
for i in range(len(CATEGORICAL_COLUMNS)):
input_i = Input(shape=(1,), dtype='int32')
dim = len(np.unique(self.all_data[CATEGORICAL_COLUMNS[i]]))
embed_dim = int(np.ceil(dim ** 0.25)) # 入力カテゴリー数の4乗根をEmbedding次元数にする
embed_i = Embedding(dim, embed_dim, input_length=1)(input_i)
flatten_i = Flatten()(embed_i)
categ_inputs.append(input_i)
categ_embeds.append(flatten_i)
# 連続的データは全結合層で一括入力
conti_input = Input(shape=(len(CONTINUOUS_COLUMNS),))
conti_dense = Dense(256, use_bias=False)(conti_input)
# 全結合層と各Embeddingの出力をくっつける
concat_embeds = concatenate([conti_dense]+categ_embeds)
concat_embeds = Activation('relu')(concat_embeds)
bn_concat = BatchNormalization()(concat_embeds)
# 更に全結合層を3層重ねる
fc1 = Dense(512, use_bias=False)(bn_concat)
ac1 = ReLU()(fc1)
bn1 = BatchNormalization()(ac1)
fc2 = Dense(256, use_bias=False)(bn1)
ac2 = ReLU()(fc2)
bn2 = BatchNormalization()(ac2)
fc3 = Dense(128)(bn2)
ac3 = ReLU()(fc3)
# 入力の層と最後の層をメンバー変数化（モデル作成用）
self.categ_inputs = categ_inputs
self.conti_input = conti_input
self.deep_component_outlayer = ac3
def wide_component(self):
# カテゴリーデータだけ線形モデルに入れる
dim = self.x_train_categ_poly.shape[1]
self.logistic_input = Input(shape=(dim,))
def create_model(self):
self.deep_component()
self.wide_component()
if self.mode == 'wide and deep':
out_layer = concatenate([self.deep_component_outlayer, self.logistic_input])
inputs = [self.conti_input] + self.categ_inputs + [self.logistic_input]
elif self.mode =='deep':
out_layer = self.deep_component_outlayer
inputs = [self.conti_input] + self.categ_inputs
else:
print('wrong mode')
return
output = Dense(1, activation='sigmoid')(out_layer)
self.model = Model(inputs=inputs, outputs=output)
def train_model(self, epochs=15, optimizer='adam', batch_size=128):
if not self.model:
print('You have to create model first')
return
if self.mode == 'wide and deep':
input_data = [self.x_train_conti] +\
[self.x_train_categ[:, i] for i in range(self.x_train_categ.shape[1])] +\
[self.x_train_categ_poly]
elif self.mode == 'deep':
input_data = [self.x_train_conti] +\
[self.x_train_categ[:, i] for i in range(self.x_train_categ.shape[1])]
else:
print('wrong mode')
return
self.model.compile(optimizer=optimizer, loss='binary_crossentropy', metrics=['accuracy'])
self.model.fit(input_data, self.y_train, epochs=epochs, batch_size=batch_size)
def evaluate_model(self):
if not self.model:
print('You have to create model first')
return
if self.mode == 'wide and deep':
input_data = [self.x_test_conti] +\
[self.x_test_categ[:, i] for i in range(self.x_test_categ.shape[1])] +\
[self.x_test_categ_poly]
elif self.mode == 'deep':
input_data = [self.x_test_conti] +\
[self.x_test_categ[:, i] for i in range(self.x_test_categ.shape[1])]
else:
print('wrong mode')
return
loss, acc = self.model.evaluate(input_data, self.y_test)
print(f'test_loss: {loss} - test_acc: {acc}')
def save_model(self, filename='wide_and_deep.h5'):
self.model.save(filename)
if __name__ == '__main__':
wide_deep_net = Wide_and_Deep()
wide_deep_net.create_model()
wide_deep_net.train_model()
wide_deep_net.evaluate_model()
wide_deep_net.save_model()
plot_model(wide_deep_net.model, to_file='model.png', show_shapes=True, show_layer_names=False)