fbprophet：预测未来的假期

我有假期的时间序列数据 . 我希望我的模型预测即将到来的假期 . 然而，似乎我的模型正在预测总体趋势，而不是在有假期的情况下分配增长 .

我有办法将未来日期指定为假期吗？我认为将它添加到假日变量中将确保它在模型/预测中 .

这是一些要测试的虚拟数据：我有2017年1月1日至今（2018年11月16日）的数据 . 每两个月有假期（销售） . 即将到来的假期（促销）是在12月 . 我想预测即将到来的12月假期 . 代码过于冗长，但希望它有助于清晰 .

%matplotlib inline
import numpy as np
import pandas as pd
from fbprophet import Prophet

times = pd.date_range('2017-01-01', '2018-11-16')
annual_cycle = np.sin(2 * np.pi * (times.dayofyear.values / 366 - 0.28))
noise = 15 * np.random.rand(annual_cycle.size)
data = 10 + (5 * annual_cycle) + noise
s = pd.Series(data, index=times)
# Add holiday data in
s.loc['2017-02-01':'2017-02-03'] = s.loc['2017-02-01':'2017-02-03'] * 3
s.loc['2017-04-01':'2017-04-03'] = s.loc['2017-04-01':'2017-04-03'] * 3
s.loc['2017-06-01':'2017-06-03'] = s.loc['2017-06-01':'2017-06-03'] * 3
s.loc['2017-08-01':'2017-08-03'] = s.loc['2017-08-01':'2017-08-03'] * 3
s.loc['2017-10-01':'2017-10-03'] = s.loc['2017-10-01':'2017-10-03'] * 3
s.loc['2017-12-01':'2017-12-03'] = s.loc['2017-12-01':'2017-12-03'] * 3
s.loc['2018-02-01':'2018-02-03'] = s.loc['2018-02-01':'2018-02-03'] * 3
s.loc['2018-04-01':'2018-04-03'] = s.loc['2018-04-01':'2018-04-03'] * 3
s.loc['2018-06-01':'2018-06-03'] = s.loc['2018-06-01':'2018-06-03'] * 3
s.loc['2018-08-01':'2018-08-03'] = s.loc['2018-08-01':'2018-08-03'] * 3
s.loc['2018-10-01':'2018-10-03'] = s.loc['2018-10-01':'2018-10-03'] * 3
# s.plot()

df = s.to_frame().reset_index()
df = df.rename(columns={"index": "ds", 0: "y"})

# Specify a holidays Series
hol1 = pd.DataFrame({'holiday': 'hol1',
                     'ds': pd.date_range(start='2/01/2017', end='2/03/2017')})
hol2 = pd.DataFrame({'holiday': 'hol2',
                     'ds': pd.date_range(start='4/01/2017', end='4/03/2017')})
hol3 = pd.DataFrame({'holiday': 'hol3',
                     'ds': pd.date_range(start='6/01/2017', end='6/03/2017')})
hol4 = pd.DataFrame({'holiday': 'hol4',
                     'ds': pd.date_range(start='8/01/2017', end='8/03/2017')})
hol5 = pd.DataFrame({'holiday': 'hol5',
                     'ds': pd.date_range(start='10/01/2017', end='10/03/2017')})
hol6 = pd.DataFrame({'holiday': 'hol6',
                     'ds': pd.date_range(start='12/01/2017', end='12/03/2017')})
hol7 = pd.DataFrame({'holiday': 'hol7',
                     'ds': pd.date_range(start='2/01/2018', end='2/03/2018')})
hol8 = pd.DataFrame({'holiday': 'hol8',
                     'ds': pd.date_range(start='4/01/2018', end='4/03/2018')})
hol9 = pd.DataFrame({'holiday': 'hol9',
                     'ds': pd.date_range(start='6/01/2018', end='6/03/2018')})
hol10 = pd.DataFrame({'holiday': 'hol10',
                      'ds': pd.date_range(start='8/01/2018', end='8/03/2018')})
hol11 = pd.DataFrame({'holiday': 'hol11',
                      'ds': pd.date_range(start='10/01/2018', end='10/03/2018')})
# Future holiday
hol12 = pd.DataFrame({'holiday': 'hol12',
                      'ds': pd.date_range(start='12/01/2018', end='12/03/2018')})
holidays = pd.concat([hol1, hol2, hol3, hol4, hol5, hol6, hol7, hol8, hol9, hol10, hol11, hol12])

m = Prophet(holidays=holidays)
m.fit(df)

# Predict up until the end of the next holiday
future = m.make_future_dataframe(periods=17)
future.tail()

forecast = m.predict(future)

fig1 = m.plot(forecast)