質問編集履歴
1
修正後のコードになります。
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -29,29 +29,32 @@
|
|
|
29
29
|
from glob import glob
|
|
30
30
|
import csv
|
|
31
31
|
import os
|
|
32
|
+
from sklearn.preprocessing import StandardScaler
|
|
32
33
|
|
|
33
34
|
def load_csv(file, y_value): # ファイル名fileのCSV読み込み，1データを複数に分割する
|
|
35
|
+
# pandas.read_csv()を使わないで書いてみる
|
|
34
36
|
_x = list() # 1ファイルのデータすべて_xに格納する
|
|
35
37
|
with open(file, "r") as f:
|
|
36
38
|
reader = csv.reader(f)
|
|
37
|
-
header = next(reader)
|
|
39
|
+
header = next(reader) # ヘッダは読み飛ばし
|
|
38
40
|
for row in reader:
|
|
39
|
-
_x.append([float(col) for col in row[:
|
|
41
|
+
_x.append([float(col) for col in row[8:25]])
|
|
40
42
|
|
|
41
43
|
# データ分割部
|
|
42
|
-
_x = np.array(_x)
|
|
44
|
+
_x = np.array(_x)
|
|
43
|
-
roll =
|
|
45
|
+
roll = 215
|
|
44
46
|
x, y = list(), list()
|
|
45
47
|
for r in range(len(_x) - roll): # 2割のデータ長さだけデータが作られる．
|
|
46
|
-
x.append(_x[r: r + roll])
|
|
48
|
+
x.append(StandardScaler().fit_transform(_x[r: r + roll].reshape(-1, 1)).reshape(_x[r: r + roll].shape))
|
|
47
49
|
y.append([y_value])
|
|
48
50
|
return np.array(x), np.array(y)
|
|
49
51
|
|
|
50
52
|
def load_data(): # 全データ読み込み
|
|
51
53
|
split = 20 # 最初の20ファイルは検証用データにする
|
|
52
54
|
x_train, y_train, x_valid, y_valid = list(), list(), list(), list()
|
|
53
|
-
for name, v in zip(["
|
|
55
|
+
for name, v in zip(["mormal", "abnormal"], [0, 1]):
|
|
54
|
-
for i, file in enumerate(glob(f"
|
|
56
|
+
for i, file in enumerate(glob(f"/content/drive/MyDrive/data/{name}/*.csv")):
|
|
57
|
+
print(file)
|
|
55
58
|
_x, _y = load_csv(file, v)
|
|
56
59
|
if i < split:
|
|
57
60
|
x_valid.extend(_x)
|
|
@@ -63,13 +66,15 @@
|
|
|
63
66
|
return map(np.array, [x_train, y_train, x_valid, y_valid])
|
|
64
67
|
|
|
65
68
|
x_train, y_train, x_valid, y_valid = load_data()
|
|
69
|
+
print(x_train.shape)
|
|
66
70
|
|
|
67
71
|
import tensorflow as tf
|
|
68
72
|
from tensorflow.python import keras
|
|
69
73
|
from keras.models import Sequential
|
|
70
|
-
from keras.layers import Conv1D, Input, Flatten, Dense
|
|
74
|
+
from keras.layers import Conv1D, Input, Flatten, Dense, AveragePooling1D, GlobalAveragePooling1D
|
|
71
75
|
from keras.optimizers import Adam
|
|
72
76
|
|
|
77
|
+
from sklearn.metrics import confusion_matrix, ConfusionMatrixDisplay
|
|
73
78
|
from sklearn.metrics import confusion_matrix
|
|
74
79
|
from sklearn.metrics import accuracy_score
|
|
75
80
|
from sklearn.metrics import precision_score
|
|
@@ -79,10 +84,12 @@
|
|
|
79
84
|
# 学習モデルを作成
|
|
80
85
|
model = Sequential()
|
|
81
86
|
model.add(Input(shape = x_train.shape[1:]))
|
|
82
|
-
model.add(Conv1D(8, 4, strides = 2, activation = "tanh"))
|
|
83
|
-
model.add(Conv1D(4,
|
|
87
|
+
model.add(Conv1D(64, 8, activation = "swish", kernel_initializer = "he_uniform"))
|
|
84
|
-
model.add(Conv1D(2, 4, strides = 2, activation = "tanh"))
|
|
85
|
-
model.add(
|
|
88
|
+
model.add(AveragePooling1D())
|
|
89
|
+
model.add(Conv1D(128, 4, activation = "swish", kernel_initializer = "he_uniform"))
|
|
90
|
+
model.add(AveragePooling1D())
|
|
91
|
+
model.add(Conv1D(256, 3, activation = "swish", kernel_initializer = "he_uniform"))
|
|
92
|
+
model.add(GlobalAveragePooling1D())
|
|
86
93
|
model.add(Dense(1, activation = "sigmoid"))
|
|
87
94
|
|
|
88
95
|
model.compile(
|
|
@@ -96,6 +103,7 @@
|
|
|
96
103
|
x_train, y_train,
|
|
97
104
|
batch_size = 8,
|
|
98
105
|
epochs = 50,
|
|
106
|
+
class_weight = {0: 4, 1: 1},
|
|
99
107
|
validation_data = (x_valid, y_valid)
|
|
100
108
|
)
|
|
101
109
|
|
|
@@ -103,7 +111,24 @@
|
|
|
103
111
|
# 0から1の 範囲 で得られるのでroundで 四捨五入して0か1にする
|
|
104
112
|
y_pred = np.round(model.predict(x_valid, batch_size = 8))
|
|
105
113
|
|
|
114
|
+
#混同行列
|
|
115
|
+
tn, fp, fn, tp = confusion_matrix(y_valid, y_pred).ravel() #混同行列のそれぞれの結果を取得
|
|
116
|
+
print("TN", tn)
|
|
117
|
+
print("FP", fp)
|
|
118
|
+
print("FN", fn)
|
|
119
|
+
print("TP", tp)
|
|
120
|
+
|
|
106
121
|
#正解率(Accuracy)
|
|
107
122
|
score = accuracy_score(y_valid, y_pred)
|
|
108
123
|
print(f"正解率：{score * 100}%")
|
|
124
|
+
|
|
125
|
+
#適合率（Precision）
|
|
126
|
+
print('Precision:', precision_score(y_valid,y_pred))
|
|
127
|
+
|
|
128
|
+
#再現率（Recall）
|
|
129
|
+
print('Recall:', recall_score(y_valid,y_pred))
|
|
130
|
+
|
|
131
|
+
#F値（F-measure）
|
|
132
|
+
print('F1 score:', f1_score(y_valid,y_pred))
|
|
133
|
+
print("Classification report")
|
|
109
134
|
```
|