Assignment5.py

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from google.colab import files
uploaded = files.upload()

# 코인데스크 사이트에서 1년치 비트코인 가격 데이터 읽기
f=open("BTC_USD_2019-02-28_2020-02-27-CoinDesk.csv","r")
coindesk_data=pd.read_csv(f,header=0)
seq=coindesk_data[['Closing Price (USD)','24h Open (USD)','24h High (USD)','24h Low (USD)']].to_numpy() # 종가, 시가, 고가, 저가를 모두 취함

# 시계열 데이터를 윈도우 단위로 자르는 함수
def seq2dataset(seq,window,horizon):
    X=[]; Y=[]
    for i in range(len(seq)-(window+horizon)+1):
        x=seq[i:(i+window)]
        y=(seq[i+window+horizon-1])
        X.append(x); Y.append(y)
    return np.array(X), np.array(Y)

w=7 # 윈도우 크기
h=1 # 수평선 계수

X,Y = seq2dataset(seq,w,h)
print(X.shape,Y.shape)
print(X[0],Y[0])

from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, LSTM, Dropout

# 훈련 집합과 테스트 집합으로 분할
split=int(len(X)*0.7)
x_train=X[0:split]; y_train=Y[0:split]
x_test=X[split:]; y_test=Y[split:]

# LSTM 모델의 설계와 학습
model = Sequential()
model.add(LSTM(units=16,activation='relu',input_shape=x_train[0].shape))
#model.add(LSTM(units=32,activation='relu',input_shape=x_train[0].shape))
#model.add(LSTM(units=64,activation='relu',input_shape=x_train[0].shape))
#model.add(LSTM(units=128,activation='relu',input_shape=x_train[0].shape))
#model.add(LSTM(units=256,activation='relu',input_shape=x_train[0].shape))
model.add(Dense(4))
model.compile(loss='mae',optimizer='adam',metrics=['mae'])
hist=model.fit(x_train,y_train,epochs=100,batch_size=1,validation_data=(x_test,y_test),verbose=2)

# LSTM 모델 평가
ev=model.evaluate(x_test,y_test,verbose=0)
print("손실 함수:",ev[0],"MAE:",ev[1])

# LSTM 모델로 예측 수행
pred=model.predict(x_test) # LSTM
print("LSTM 평균절댓값백분율오차(MAPE):",sum(abs(y_test-pred)/y_test)/len(x_test))

# 학습 곡선
plt.plot(hist.history['mae'])
plt.plot(hist.history['val_mae'])
plt.title('Model mae')
plt.ylabel('mae')
plt.xlabel('Epoch')
plt.ylim([100,600])
plt.legend(['Train','Validation'], loc='best')
plt.grid()
plt.show()