시계열 수치입력 예측모델 레시피(상태유지 순환신경망 모델)

# 1. 사용할 패키지 불러오기

import numpy as np

from keras.models import Sequential

from keras.layers import Dense, LSTM, Dropout

from sklearn.preprocessing import MinMaxScaler

import matplotlib.pyplot as plt

%matplotlib inline

# 2. 함수와 클래스 만들기

def create_dataset(signal_data, look_back=1):

    dataX, dataY = [], []

    for i in range(len(signal_data)-look_back):

        dataX.append(signal_data[i:(i+look_back), 0])

        dataY.append(signal_data[i + look_back, 0])

    return np.array(dataX), np.array(dataY)

class CustomHistory(keras.callbacks.Callback):

    def init(self):

        self.train_loss = []

        self.val_loss = []

        

    def on_epoch_end(self, batch, logs={}):

        self.train_loss.append(logs.get('loss'))

        self.val_loss.append(logs.get('val_loss'))

look_back = 40

# 3. 데이터 핸들링

signal_data = np.cos(np.arange(1600)*(20*np.pi/1000))[:, None]

# 3.1 데이터 전처리

scaler = MinMaxScaler(feature_range = (0, 1))

signal_data = scaler.fit_transform(signal_data)

# 3.2 데이터 분리

train = signal_data[0:800]

val   = signal_data[800:1200]

test  = signal_data[1200:]

# 3.3 데이터 생성

x_train, y_train = create_dataset(train, look_back)

x_val, y_val     = create_dataset(val, look_back)

x_test, y_test   = create_dataset(val, look_back)

# 3.4 데이터 전처리

x_train = np.reshape(x_train, (x_train.shape[0], x_train.shape[1], 1))

x_val   = np.reshape(x_val, (x_val.shape[0], x_val.shape[1], 1))

x_test  = np.reshape(x_test, (x_test.shape[0], x_test.shape[1], 1))

# 4. 모델 구성하기

model = Sequential()

model.add(LSTM(32, batch_input_shape = (1, look_back, 1), stateful = True))

model.add(Dropout(0.3))

model.add(Dense(1))

# 5. 모델 학습과정 설정하기
model.compile(loss='mean_squared_error', optimizer='adam')

# 6. 모델 학습시키기 

custom_hist = CustomHistory()

custom_hist.init()

for i in range(200):

    model.fit(x_train, y_train, epochs=1, batch_size=1, shuffle = False, callbacks = [custom_hist], validation_data=(x_val, y_val))

    model.reset_states()

# 7. 학습과정 살펴보기

plt.plot(custom_hist.train_loss)

plt.plot(custom_hist.val_loss)

plt.ylim(0.0, 0.15)

plt.ylabel('loss')

plt.xlabel('epoch')

plt.legend(['train', 'val'], loc='upper left')

plt.show()

# 8. 모델 사용하기

trainScore = model.evaluate(x_train, y_train, batch_size=1, verbose=0)

model.reset_states()

print('Train Score: ', trainScore)

valScore = model.evaluate(x_val, y_val, batch_size=1, verbose=0)

model.reset_states()

print('Validataion Score: ', valScore)

testScore = model.evaluate(x_test, y_test, batch_size=1, verbose=0)

model.reset_states()

print('Test Score: ', testScore)

# 9. 모델 사용하기

look_ahead = 250

xhat = x_test[0]

predictions = np.zeros((look_ahead,1))

for i in range(look_ahead):

    prediction = model.predict(np.array([xhat]), batch_size=1)

    predictions[i] = prediction

    xhat = np.vstack([xhat[1:],prediction])

    

plt.figure(figsize=(12,5))

plt.plot(np.arange(look_ahead),predictions,'r',label="prediction")

plt.plot(np.arange(look_ahead),y_test[:look_ahead],label="test function")

plt.legend()

plt.show()

[출처] 블록과 함께하는 파이썬 딥러닝 케라스, 김태영 지음, DigitalBooks, p263~265