質問編集履歴
1
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -9,3 +9,93 @@
|
|
|
9
9
|
説明変数の数やデータ数を増やすことは効果があるでしょうか？深層学習を勉強し始めたばかりの素人なのですが、何か予測精度を向上する方法をご存知でしたらご教示頂けますと幸いです。
|
|
10
10
|
|
|
11
11
|
どうぞよろしくお願い申し上げます。
|
|
12
|
+
|
|
13
|
+
|
|
14
|
+
|
|
15
|
+
なお、keras のコードは以下になります。
|
|
16
|
+
|
|
17
|
+
|
|
18
|
+
|
|
19
|
+
df = pd.read_csv('./data.csv')
|
|
20
|
+
|
|
21
|
+
x = df.loc[:, 'f_' : 'g_']
|
|
22
|
+
|
|
23
|
+
y = df.loc[:, 'x_' : 'y_']
|
|
24
|
+
|
|
25
|
+
|
|
26
|
+
|
|
27
|
+
x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.2, shuffle= False)
|
|
28
|
+
|
|
29
|
+
|
|
30
|
+
|
|
31
|
+
x_train_mean = x_train.mean(axis=0) # 正規化
|
|
32
|
+
|
|
33
|
+
x_train_std = x_train.std(axis=0)
|
|
34
|
+
|
|
35
|
+
x_train -= x_train_mean
|
|
36
|
+
|
|
37
|
+
x_train /= x_train_std
|
|
38
|
+
|
|
39
|
+
y_train_mean = y_train.mean()
|
|
40
|
+
|
|
41
|
+
y_train_std = y_train.std()
|
|
42
|
+
|
|
43
|
+
y_train -= y_train_mean
|
|
44
|
+
|
|
45
|
+
y_train /= y_train_std
|
|
46
|
+
|
|
47
|
+
x_test -= x_train_mean
|
|
48
|
+
|
|
49
|
+
x_test /= x_train_std
|
|
50
|
+
|
|
51
|
+
y_test -= y_train_mean
|
|
52
|
+
|
|
53
|
+
y_test /= y_train_std
|
|
54
|
+
|
|
55
|
+
|
|
56
|
+
|
|
57
|
+
from keras.models import Sequential
|
|
58
|
+
|
|
59
|
+
from keras.layers import Dense, Activation
|
|
60
|
+
|
|
61
|
+
model = Sequential()
|
|
62
|
+
|
|
63
|
+
model.add(Dense(32, activation='relu', input_shape=(x_train.shape[1],))) # 入力層
|
|
64
|
+
|
|
65
|
+
model.add(Dense(32, activation='relu')) # 隠れ層(5層)
|
|
66
|
+
|
|
67
|
+
model.add(Dense(32, activation='relu'))
|
|
68
|
+
|
|
69
|
+
model.add(Dense(32, activation='relu'))
|
|
70
|
+
|
|
71
|
+
model.add(Dense(32, activation='relu'))
|
|
72
|
+
|
|
73
|
+
model.add(Dense(32, activation='relu'))
|
|
74
|
+
|
|
75
|
+
model.add(Dense(2)) # 出力層
|
|
76
|
+
|
|
77
|
+
|
|
78
|
+
|
|
79
|
+
from keras.optimizers import adam
|
|
80
|
+
|
|
81
|
+
model.compile(optimizer = 'adam', # 最適化アルゴリズム: Adam
|
|
82
|
+
|
|
83
|
+
loss = 'mse', # 損失関数: mse（平均二乗誤差）
|
|
84
|
+
|
|
85
|
+
metrics = ['mae']) # 評価関数: mae（平均絶対誤差）
|
|
86
|
+
|
|
87
|
+
|
|
88
|
+
|
|
89
|
+
history = model.fit(x_train, y_train, # トレーニングデータ
|
|
90
|
+
|
|
91
|
+
batch_size = 1, # バッチサイズ
|
|
92
|
+
|
|
93
|
+
epochs = 100, # エポック数
|
|
94
|
+
|
|
95
|
+
verbose = 1, # ログ出力の指定
|
|
96
|
+
|
|
97
|
+
validation_data = (x_test, y_test)) # テストデータ
|
|
98
|
+
|
|
99
|
+
|
|
100
|
+
|
|
101
|
+
model.predict(x_test)
|