[Python LSTM ]LSTMのモデルで2値分類を実行する

         fTEpoch    fLen   fCLen    pDelta  ipProt1  ipProt6  ipProt17  portSrc  portDst
0       1.541251e+09  1373.0  1373.0  0.550714        0        0         1    46649     5087
1       1.541251e+09  1514.0  1514.0  0.551437        0        0         1    46657    39736
2       1.541251e+09  1498.0  1498.0  0.549570        0        0         1    46657    39736
3       1.541251e+09  1514.0  1514.0  0.549820        0        0         1    46667    46528
4       1.541251e+09   972.0   972.0  0.549818        0        0         1    46667    46528
...              ...     ...     ...       ...      ...      ...       ...      ...      ...
199995  1.609564e+09    60.0    42.0  0.413288        0        0         1     1526     5897
199996  1.609564e+09    60.0    42.0  0.413286        0        0         1     1526     7341
199997  1.609564e+09    98.0    66.0  0.546451        0        1         0    27683    16172
199998  1.609564e+09    67.0    66.0  0.545296        0        1         0    27683    16172
199999  1.609564e+09   113.0    66.0  0.547878        0        1         0    27683    16172

# ライブラリのインポート
import numpy as np
import pandas as pd
from matplotlib import pyplot as plt
import time
import pickle
from  settingfile import BASE_PATH
from sklearn.preprocessing import LabelEncoder

# ルックバック数
look_back = 6 # 偶数にする

# データ数
sample_size = len(dFTrain_oc) - look_back

# 予測に用いる期間
test_size = int(len(dFTest_oc))

# データセットを作る関数
def make_dataset(raw_data, look_back=1):
    _X = []
    _y = []

    for i in range(len(raw_data) - look_back):
        _X.append(raw_data[i : i + look_back])
        _y.append(raw_data[i + look_back])
    _X = np.array(_X).reshape(len(_X), look_back, 1)
    _y = np.array(_y).reshape(len(_y), 1)

    return _X, _y

from sklearn import preprocessing

columns = list(dFTrain_oc.columns)
print(columns)

# 最小0、最大1に正規化
Xs = []
for i in range(len(columns)):
    Xs.append(preprocessing.minmax_scale(dFTrain_oc[columns[i]]))
Xs = np.array(Xs)

Xs_t = []
for i in range(len(columns)):
    Xs_t.append(preprocessing.minmax_scale(dFTest_oc[columns[i]]))
Xs_t = np.array(Xs_t)

# 各数値データを作成
X_1, y_1 = make_dataset(Xs[0], look_back=look_back)
X_2, y_2 = make_dataset(Xs[1], look_back=look_back)
X_3, y_3 = make_dataset(Xs[2], look_back=look_back)
X_4, y_4 = make_dataset(Xs[3], look_back=look_back)
X_5, y_5 = make_dataset(Xs[4], look_back=look_back)
X_6, y_6 = make_dataset(Xs[5], look_back=look_back)
X_7, y_7 = make_dataset(Xs[6], look_back=look_back)
X_8, y_8 = make_dataset(Xs[7], look_back=look_back)
X_9, y_9 = make_dataset(Xs[8], look_back=look_back)
X_1_t, y_1_t = make_dataset(Xs_t[0], look_back=look_back)
X_2_t, y_2_t = make_dataset(Xs_t[1], look_back=look_back)
X_3_t, y_3_t = make_dataset(Xs_t[2], look_back=look_back)
X_4_t, y_4_t = make_dataset(Xs_t[3], look_back=look_back)
X_5_t, y_5_t = make_dataset(Xs_t[4], look_back=look_back)
X_6_t, y_6_t = make_dataset(Xs_t[5], look_back=look_back)
X_7_t, y_7_t = make_dataset(Xs_t[6], look_back=look_back)
X_8_t, y_8_t = make_dataset(Xs_t[7], look_back=look_back)
X_9_t, y_9_t = make_dataset(Xs_t[8], look_back=look_back)

# 多変量LSTMに対応するために各データを結合
X_con = np.concatenate([X_2,X_3,X_4,X_5,X_6,X_7], axis=2)
X_con_t = np.concatenate([X_2_t,X_3_t,X_4_t,X_5_t,X_6_t,X_7_t], axis=2)

# x_test,x_train,y_test,y_trainの設定====================
X = X_con
y = y_3
X_t = X_con_t
y_t = y_3_t

# 訓練データを定義
X_train = X # X_con
y_train = y
X_test = X_t
y_test = y_t

import tensorflow as tf
from tensorflow.keras.layers import Input, LSTM, Dense, BatchNormalization
from tensorflow.keras.models import Model
from tensorflow.keras.optimizers import SGD, Adam

# LSTMモデルを作成する関数
def create_LSTM_model():
    input = Input(shape=(np.array(X_train).shape[1], np.array(X_train).shape[2]))
    x = LSTM(64, return_sequences=True)(input)
    x = BatchNormalization()(x)
    x = LSTM(64)(x)
    output = Dense(1, activation='tanh')(x)
    model = Model(input, output)
    return model

model = create_LSTM_model()
model.summary()
model.compile(optimizer=Adam(learning_rate=0.0001), loss='mean_squared_error')
history = model.fit(X_train, y_train, epochs=10, batch_size=64, verbose=1, validation_data=(X_test, y_test))

loss = history.history["loss"]
val_loss = history.history["val_loss"]

epochs = range(len(loss))

plt.figure(figsize=(15,5))
plt.plot(epochs, loss, "b", label="Training loss")
plt.plot(epochs, val_loss, "r", label="Validatioin loss")
plt.xlabel("epoch")
plt.ylabel("loss")
plt.title("Training and validation loss")
plt.legend(fontsize=20)
plt.show()

predictions = model.predict(X_test)

predictions_round = np.round(predictions)

plt.figure(figsize=(15,5))
plt.plot( predictions_round, "r", label="predictions")
plt.legend(fontsize=20)
plt.show()

from sklearn.metrics import confusion_matrix

sizePos = 100000 - int(look_back / 2)#異常値のデータ数
sizeNeg = 100000 - int(look_back / 2) #正常値のデータ数
calc_oc = calc([1] * sizePos + [0] * sizeNeg, predictions_round, [1, 0]) # 正常値は1,異常値は0で表現する配列を第１引数に渡している？？

tp, fn, fp, tn = confusion_matrix(a, b, labels=l).ravel() # 第１引数に正解ラベル、第２引数に予測ラベル