[시계열 데이터 분석] numpy를 이용한 시계열 데이터 분석

 

numpy를 이용한 시계열 데이터 분석¶

In [5]:

import warnings
warnings.filterwarnings('ignore')
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline
from pandas_datareader import data
from fbprophet import Prophet
from datetime import datetime

 

폰트 설정¶

In [6]:

from matplotlib import font_manager, rc

rc('font', family='AppleGothic')
plt.rcParams['axes.unicode_minus'] = False

In [37]:

pinkwink_web = pd.read_csv('./data/07. PinkWink Web Traffic.csv', encoding='utf-8', thousands=',', names=['date', 'hit'], index_col=0)
pinkwink_web = pinkwink_web[pinkwink_web['hit'].notnull()]
pinkwink_web.head(3)

Out[37]:

	hit
date
16. 7. 1.	766.0
16. 7. 2.	377.0
16. 7. 3.	427.0

In [38]:

pinkwink_web['hit'].plot(figsize=(12, 4), grid=True)

Out[38]:

<matplotlib.axes._subplots.AxesSubplot at 0x117a161d0>

 

In [13]:

time = np.arange(0, len(pinkwink_web))
traffic = pinkwink_web['hit'].values
fx = np.linspace(0, time[-1], 1000)

In [19]:

def error(f, x, y):
    return np.sqrt(np.mean((f(x)-y)**2))

In [31]:

fp1 = np.polyfit(time, traffic, 1)
f1 = np.poly1d(fp1)

f2p = np.polyfit(time, traffic, 2)
f2 = np.poly1d(f2p)

f3p = np.polyfit(time, traffic, 3)
f3 = np.poly1d(f3p)

f5p = np.polyfit(time, traffic, 5)
f5 = np.poly1d(f5p)

f15p = np.polyfit(time, traffic, 15)
f15 = np.poly1d(f15p)

print(error(f1, time, traffic))
print(error(f2, time, traffic))
print(error(f3, time, traffic))
print(error(f5, time, traffic))
print(error(f15, time, traffic))

 

430.8597308110963
430.62841018946955
429.5328046676293
389.8483612021826
330.4777307248983

In [32]:

plt.figure(figsize=(10, 6))
plt.scatter(time, traffic, s=10)

plt.plot(fx, f1(fx), lw=4, label='f1')
plt.plot(fx, f2(fx), lw=4, label='f2')
plt.plot(fx, f3(fx), lw=4, label='f3')
plt.plot(fx, f5(fx), lw=4, label='f5')
plt.plot(fx, f15(fx), lw=4, label='f15')
plt.grid(True, linestyle='-', color='0.75')
plt.legend(loc=2) # f1, f2 ...위치
plt.show()

 

 

Prophet 모듈을 이용한 forecast 예측¶

In [39]:

df = pd.DataFrame({'ds': pinkwink_web.index, 'y': pinkwink_web['hit']})
df.reset_index(inplace=True)
df['ds'] = pd.to_datetime(df['ds'], format="%y. %m. %d.")
del df['date']
df.head(3)

Out[39]:

	ds	y
0	2016-07-01	766.0
1	2016-07-02	377.0
2	2016-07-03	427.0

In [40]:

m = Prophet(yearly_seasonality=True)
m.fit(df)

Out[40]:

<fbprophet.forecaster.Prophet at 0x117a06c50>

In [41]:

future = m.make_future_dataframe(periods=60)
future.tail(3)

Out[41]:

	ds
422	2017-08-27
423	2017-08-28
424	2017-08-29

In [43]:

forecast = m.predict(future)
forecast[['ds', 'yhat', 'yhat_lower', 'yhat_upper']].tail(3)

Out[43]:

	ds	yhat	yhat_lower	yhat_upper
422	2017-08-27	641.503999	435.113740	849.683688
423	2017-08-28	1194.071095	979.746565	1394.709571
424	2017-08-29	1229.894880	1008.136578	1424.921642

In [44]:

m.plot(forecast)

Out[44]:

 

In [46]:

m.plot_components(forecast)

Out[46]: