【Pythonで株価予測】アップル（Apple）の今後の株価は？ （複数期間で検証）

import pandas as pd
import pandas_datareader.data as web
import matplotlib.pyplot as plt
import numpy as np
from fbprophet import Prophet
from sklearn.metrics import mean_squared_log_error

################################################################
st = '2016/01/01'
ed = '2022/06/17'
st_train = '2016/01/01'
ed_train = '2022/04/30'
st_test = ed_train
ed_test = '2022/05/31'
period = 61 #予測日数
y_type =1 #0:オリジナルの株価、1:対数株価
changepoint_flag = 0 #0:changepointを考慮しない、1:FOMCの翌日
fomc_list = ['2019-01-31','2019-03-21','2019-05-02','2019-06-20',
             '2019-08-01','2019-09-19','2019-11-01','2019-12-12',
             '2020-01-30','2020-03-19','2020-04-30','2020-06-11',
             '2020-07-30','2020-09-17','2020-11-06','2020-12-17',
             '2021-01-28','2021-03-18', '2021-04-29','2021-06-17',
             '2021-07-29','2021-09-23','2021-11-04','2021-12-16',
             '2022-01-27','2022-03-17','2022-05-05']
################################################################

main_target = 'AAPL'
lst1= ['AAPL']
stooq = web.DataReader(lst1, 'yahoo',start=st,end=ed)

df = stooq[st_train:ed_train]

#①株価 or ②③株価の対数
if y_type == 1:
   df = df.dropna(how='any')
   df['Adj Close'] = np.log(df['Adj Close'])
   df = df.reset_index()
   df[['ds','y']] = df[['Date' ,'Adj Close']]
   df = df[['ds','y']]
else:
   df = df.reset_index()
   df[['ds','y']] = df[['Date' ,'Adj Close']]
   df = df[['ds','y']]

df_actual = stooq[st_test:ed_test]
if y_type == 1:
   df_actual = df_actual.dropna(how='any')
   df_actual['Adj Close'] = np.log(df_actual['Adj Close'])
df_actual = df_actual.reset_index()
df_actual[['ds','y']] = df_actual[['Date' ,'Adj Close']]
df_actual = df_actual[['ds','y']]


if y_type == 1:
   stooq = stooq.dropna(how='any')
   stooq['Adj Close'] = np.log(stooq['Adj Close'])
stooq = stooq.reset_index()
stooq[['ds','y']] = stooq[['Date' ,'Adj Close']]
stooq = stooq[['ds','y']]


#FOMC日程考慮
if changepoint_flag==0:
   m = Prophet(weekly_seasonality=True)
elif changepoint_flag==1:
   m = Prophet(weekly_seasonality=True,changepoints=fomc_list)


#モデル学習+予測
m.fit(df)
future = m.make_future_dataframe(periods=period)
forecast = m.predict(future)
plt.rcParams["font.size"] = "25"
fig, ax = plt.subplots(figsize=(20,10))
pd.plotting.register_matplotlib_converters()
m.plot(forecast,ax=ax);
ax.plot(stooq["ds"],stooq["y"],color="r")
plt.rcParams["font.size"] = "13.3"
ax.set_title(str(main_target))
m.plot_components(forecast);

stooq.columns = stooq.columns.droplevel(1)
ans = pd.merge(forecast, stooq, on="ds", how='inner')
ans.set_index("ds",inplace=True)

y_pred = ans[st_test:ed_test]["yhat"]
y_true = ans[st_test:ed_test]["y"]

if y_type == 1:
   y_pred = np.exp(y_pred)
   y_true = np.exp(y_true)

#評価指標1：RMSLE
RMSLE=np.sqrt(mean_squared_log_error (y_true, y_pred))
print("RMSLE:",RMSLE)

y_pred_lower = ans[st_test:ed_test]["yhat_lower"]
y_pred_upper = ans[st_test:ed_test]["yhat_upper"]
ans_judge = ans[st_test:ed_test].copy()


if y_type == 1:
   y_pred_lower = np.exp(y_pred_lower)
   y_pred_upper = np.exp(y_pred_upper)


#評価指標2
cnt = 0
for i in ans_judge.index:
   if y_pred_lower[str(i)] <= y_true[str(i)] <= y_pred_upper[str(i)]:
      cnt += 1
print("予測数：",len(ans_judge.index),"範囲内：",cnt)