[신경망] RNN을 이용한 주식가격 예측 알고리즘 코드

 

데이터 

 

data-02-stock_daily.csv

0.04MB

 

코드 

import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
import seaborn as sns

def MinMaxScaler(data):
    denom=np.max(data,0)-np.min(data,0)
    nume=data-np.min(data,0)
    return nume/denom    
    #정규화
    
path = "C:\\Users\\student\\Desktop\\DY\\★ 데이터\\105. deep-learning-dataset\\"

xy=np.loadtxt(path+"data-02-stock_daily.csv", delimiter=",", skiprows=0+1+1)
xy=xy[::-1] #역순으로 저장

# skip multiple rows 여러 줄 skip하는 방법: +로 int 를 더해준다 

 

seqLength=7 #window size
dataDim=5 #시가,종가,...거래량 5개 컬럼
hiddenDim=10
outputDim=1
lr=0.01
iterations=500

trainSize=int(len(xy)*0.7)
trainSet=xy[0:trainSize] #0:499
testSet=xy[trainSize-seqLength:]#493:499->500

trainSet=MinMaxScaler(trainSet)
testSet=MinMaxScaler(testSet)

def buildDataSet(timeSeries, seqLength):
    xdata=[]
    ydata=[]
    for i in range(0, len(timeSeries)-seqLength):
        tx=timeSeries[i:i+seqLength,:]
        ty=timeSeries[i+seqLength,[-1]]
        xdata.append(tx)
        ydata.append(ty)
    return np.array(xdata), np.array(ydata)
    
trainX, trainY=buildDataSet(trainSet, seqLength)
testX, testY=buildDataSet(testSet, seqLength)

학습

loss= tf.reduce_mean(tf.square(yhat-y))
opt=tf.train.AdamOptimizer(lr)
train=opt.minimize(loss)

#rmse
targets=tf.placeholder(tf.float32, [None,1])
predictions=tf.placeholder(tf.float32, [None,1])
rmse=tf.sqrt(tf.reduce_mean(tf.square(targets-predictions)))

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    for i in range(iterations):
        _,cv=sess.run([train, loss], feed_dict={
            x:trainX, y:trainY
        })
        print("step:{} loss:{}".format(i, cv))
    testPredict=sess.run(yhat, feed_dict={x:testX})
    rmsev=sess.run(rmse, feed_dict={
        targets:testY,
        predictions:testPredict
    })
    print("rmse value: {}".format(rmsev))
    
    

테스트

trainX.shape #(505,7,5)
#505개 샘플, 7(일주일), 5(5개 컬럼)
testX.shape #(220, 7, 5)

x=tf.placeholder(tf.float32, [None, seqLength, dataDim])
y=tf.placeholder(tf.float32, [None, 1])

cell=tf.contrib.rnn.BasicLSTMCell(
    num_units=hiddenDim, 
    state_is_tuple=True,
    activation=tf.tanh
    #([1,2],[3,4])
    #is_Tuple=True => 튜플 = ([1,2],[3,4])
    #         False => 텐서 = [1,2,3,4]
)

outputs, _states=tf.nn.dynamic_rnn(cell, x, dtype=tf.float32)
yhat=tf.contrib.layers.fully_connected(outputs[:,-1], outputDim, activation_fn=None)

 

%matplotlib notebook
plt.plot(testY)
plt.plot(testPredict)
plt.show()

 

유사 상세 포스팅 참고 

https://mybeta.tistory.com/27

RNN을 이용한 stock 종가 예측