prophet으로 삼성 주식 예측하기

!pip install finance-datareader

Collecting finance-datareader
  Downloading finance_datareader-0.9.31-py3-none-any.whl (17 kB)
Requirement already satisfied: pandas>=0.19.2 in /usr/local/lib/python3.7/dist-packages (from finance-datareader) (1.3.5)
Requirement already satisfied: requests>=2.3.0 in /usr/local/lib/python3.7/dist-packages (from finance-datareader) (2.23.0)
Collecting requests-file
  Downloading requests_file-1.5.1-py2.py3-none-any.whl (3.7 kB)
Requirement already satisfied: lxml in /usr/local/lib/python3.7/dist-packages (from finance-datareader) (4.2.6)
Requirement already satisfied: tqdm in /usr/local/lib/python3.7/dist-packages (from finance-datareader) (4.62.3)
Requirement already satisfied: numpy>=1.17.3 in /usr/local/lib/python3.7/dist-packages (from pandas>=0.19.2->finance-datareader) (1.21.5)
Requirement already satisfied: python-dateutil>=2.7.3 in /usr/local/lib/python3.7/dist-packages (from pandas>=0.19.2->finance-datareader) (2.8.2)
Requirement already satisfied: pytz>=2017.3 in /usr/local/lib/python3.7/dist-packages (from pandas>=0.19.2->finance-datareader) (2018.9)
Requirement already satisfied: six>=1.5 in /usr/local/lib/python3.7/dist-packages (from python-dateutil>=2.7.3->pandas>=0.19.2->finance-datareader) (1.15.0)
Requirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.7/dist-packages (from requests>=2.3.0->finance-datareader) (2021.10.8)
Requirement already satisfied: idna<3,>=2.5 in /usr/local/lib/python3.7/dist-packages (from requests>=2.3.0->finance-datareader) (2.10)
Requirement already satisfied: chardet<4,>=3.0.2 in /usr/local/lib/python3.7/dist-packages (from requests>=2.3.0->finance-datareader) (3.0.4)
Requirement already satisfied: urllib3!=1.25.0,!=1.25.1,<1.26,>=1.21.1 in /usr/local/lib/python3.7/dist-packages (from requests>=2.3.0->finance-datareader) (1.24.3)
Installing collected packages: requests-file, finance-datareader
Successfully installed finance-datareader-0.9.31 requests-file-1.5.1

import pandas as pd
import FinanceDataReader as fdr
import matplotlib.pyplot as plt

df = fdr.DataReader('005930','2018-01-01','2022-02-25')   # 삼성전자: 005930
df

	Open	High	Low	Close	Volume	Change
Date
2018-01-02	51380	51400	50780	51020	169485	0.001177
2018-01-03	52540	52560	51420	51620	200270	0.011760
2018-01-04	52120	52180	50640	51080	233909	-0.010461
2018-01-05	51300	52120	51200	52120	189623	0.020360
2018-01-08	52400	52520	51500	52020	167673	-0.001919
...	...	...	...	...	...	...
2022-02-21	73200	74300	72600	74200	10489717	-0.001346
2022-02-22	73000	73400	72800	73400	11692469	-0.010782
2022-02-23	73800	73800	72800	73000	10397964	-0.005450
2022-02-24	72300	72300	71300	71500	15759283	-0.020548
2022-02-25	72100	72600	71900	71900	13062251	0.005594

1023 rows × 6 columns

data = df['Close'].values

plt.figure(figsize=(12, 8))
plt.plot(data)
plt.show()

from sklearn.preprocessing import MinMaxScaler
import numpy as np

data = np.array(data).reshape(-1,1)    # 표준화를 하기 위해 사이즈를 (-1,1)로 조정

scaler = MinMaxScaler(feature_range=(0, 1))
scaled = scaler.fit_transform(data)
scaled

array([[0.25340803],
       [0.26461251],
       [0.25452848],
       ...,
       [0.66386555],
       [0.63585434],
       [0.643324  ]])

test_idx = int(len(scaled) * 0.8)

train = scaled[:test_idx]
test = scaled[test_idx:]

페이스북 Prophet

Prophet은 한마디로 요약하면, 시계열 데이터를 모델링하기 위한 파이썬 모듈

트렌드와 주기적 특성뿐 아니라 예외적이고 이벤트와 같은 휴가철 상황까지도 모델링이 됨

df['ds'] = pd.to_datetime(df.index)
df['y'] = df['Close']

df[['ds', 'y']].iloc[:-10]

	ds	y
Date
2018-01-02	2018-01-02	51020
2018-01-03	2018-01-03	51620
2018-01-04	2018-01-04	51080
2018-01-05	2018-01-05	52120
2018-01-08	2018-01-08	52020
...	...	...
2022-02-07	2022-02-07	73000
2022-02-08	2022-02-08	73500
2022-02-09	2022-02-09	74700
2022-02-10	2022-02-10	75400
2022-02-11	2022-02-11	74900

1013 rows × 2 columns

# prophet 모듈 세팅
from fbprophet import Prophet

# Day 단위로 데이터가 구성되어 있으므로, 일 단위 주기성 활성화
model = Prophet(daily_seasonality=True) 

# 데이터 학습 시작 -> 기계학습
model.fit(df[['ds', 'y']].iloc[:-10])

# 주가 예측 위한 날짜 데이터 세팅 -> 기존 데이터 + 향후 14일치 예측값
future = model.make_future_dataframe(periods=10)

# 주가 예측
forecast = model.predict(future)

# forecast.columns ->
'''
  Index(['ds', 'trend', 'yhat_lower', 'yhat_upper', 'trend_lower', 'trend_upper',
       'additive_terms', 'additive_terms_lower', 'additive_terms_upper',
       'daily', 'daily_lower', 'daily_upper', 'weekly', 'weekly_lower',
       'weekly_upper', 'yearly', 'yearly_lower', 'yearly_upper',
       'multiplicative_terms', 'multiplicative_terms_lower',
       'multiplicative_terms_upper', 'yhat'],
      dtype='object')
''' 

# 필요한 컬럼만 보기
forecast[['ds', 'yhat', 'yhat_lower', 'yhat_upper']].tail(11)

# 모델이 제공하는 시각화
model.plot(forecast)

plt.subplot(1,2,1)
plt.plot(forecast.tail(10)['yhat'])
plt.title('preds')
plt.subplot(1,2,2)
plt.title('label')
plt.plot(df.tail(10)['Close'])
plt.show()

기준 별 주가 변동 파악하기

시계열의 추세, 연간 계절성, 주간 계절성이 표시됨

model.plot_components(forecast)

model = Prophet() 

df['y'] = df['Change']
# 데이터 학습 시작 -> 기계학습
model.fit(df[['ds', 'y']].iloc[:-5])

# 주가 예측 위한 날짜 데이터 세팅 -> 기존 데이터 + 향후 10일치 예측값
future = model.make_future_dataframe(periods=10)

forecast = model.predict(future)

target_changes = forecast['yhat'].tail(11)

# 모델이 제공하는 시각화
model.plot(forecast)

INFO:fbprophet:Disabling daily seasonality. Run prophet with daily_seasonality=True to override this.

target_mean_change = round(target_changes.mean(),4)
target_mean_change = 1 - abs(target_mean_change)
target_mean_change

0.9989

preds = []

preds.append(df.iloc[-1]['Close'] * target_mean_change)
preds.append(df.iloc[-1]['Close'] * target_mean_change * target_mean_change)
preds.append(df.iloc[-1]['Close'] * target_mean_change * target_mean_change * target_mean_change) 
preds.append(df.iloc[-1]['Close'] * target_mean_change * target_mean_change * target_mean_change * target_mean_change)

plt.subplot(1,2,1)
plt.title('preds')z
plt.plot(preds)
plt.subplot(1,2,2)
plt.title('label')
plt.plot(df.tail(5)['Close'])
plt.show()