价格预测对未来一年有效，但对明天无效

Question

我使用facebook Prophet预测库对未来价格进行建模，我可以获得明年的所有数据，但明天的价格会停滞不前。

这是预测代码：

m = Prophet(
seasonality_mode="multiplicative",
yearly_seasonality=True
)
m.fit(df_crypto_market_chart_exploded)
future = m.make_future_dataframe(periods = 365)
forecast = m.predict(future)
forecast_result_tail = forecast[['ds', 'yhat', 'yhat_lower', 'yhat_upper']].tail()
#print('n--- selected: FORECAST RESULT TAIL ---')
#print(forecast_result_tail)
forecast_result_head = forecast[['ds', 'yhat', 'yhat_lower', 'yhat_upper']].head()
print('n--- selected: FORECAST RESULT HEAD ---')
print(forecast_result_head)
next_day = (datetime.today() + timedelta(days=3)).strftime('%Y-%m-%d')
print('n--- selected: FORECAST DS TYPE ---')
print(forecast['ds'].dtype)
tomorrow_result = forecast['ds'].dt.date == datetime.today() + timedelta(days=1)
print('n--- selected: TOMORROW RESULTS ---')
print(tomorrow_result.head())
forecast[tomorrow_result]['yhat'].item()

**编辑-**我根据建议的解决方案修改了代码，但我仍然遇到下面列出的numpy错误

--- selected: FORECAST RESULT HEAD ---
ds         yhat  yhat_lower   yhat_upper
0 2017-06-02 00:00:00.000  2359.464229  605.624033  4037.385324
1 2017-06-03 00:00:00.000  2371.962255  661.690821  4155.470592
2 2017-06-04 00:00:00.000  2382.640816  683.686292  4249.680911
3 2017-06-05 00:00:00.000  2387.433948  570.835239  4240.934785
4 2017-06-06 23:29:55.701  2226.178108  344.391258  4062.279165

--- selected: FORECAST DS TYPE ---
datetime64[ns]

--- selected: TOMORROW RESULTS ---
0    False
1    False
2    False
3    False
4    False
Name: ds, dtype: bool
Traceback (most recent call last):
File "d:/Projects/price_prediction.py", line 69, in <module>
forecast[tomorrow_result]['yhat'].item()
File "C:Usersabcanaconda3envsvenvlibsite-packagespandascorebase.py", line 331, in item
raise ValueError("can only convert an array of size 1 to a Python scalar")
ValueError: can only convert an array of size 1 to a Python scalar

Answer 1

pandas.dataframe可以将ds保留为datetime对象，并且还可以比较hours, minutes, seconds。

使用.strftime()也不起作用，因为它很可能不会比较datetime和string，但如果它可以比较它们，那么它仍然会使用hours, minutes, seconds作为ds中的值。

您可能需要将.dt.date用于dataframe，将.date()用于单个datetime以仅比较日期。

---

最小示例

import datetime
import time
one_day = datetime.timedelta(days=1)
tomorrow_1 = datetime.datetime.today() + one_day
time.sleep(2) # to create values with different seconds
tomorrow_2 = datetime.datetime.today() + one_day
print('tomorrow_1:', tomorrow_1, '| strftime:', tomorrow_1.strftime('%Y-%m-%d'))
print('tomorrow_2:', tomorrow_2, '| strftime:', tomorrow_2.strftime('%Y-%m-%d'))
print('           dt == dt            :', tomorrow_1 == tomorrow_2)
print('           dt == dt.strftime() :', tomorrow_1 == tomorrow_2.strftime('%Y-%m-%d'))
print('dt.strftime() == dt.strftime() :', tomorrow_1.strftime('%Y-%m-%d') == tomorrow_2.strftime('%Y-%m-%d'))
print('    dt.date() == dt.date()     :', tomorrow_1.date() == tomorrow_2.date())
# ------------------------------------------
import pandas as pd
df = pd.DataFrame({
'ds': [tomorrow_1, tomorrow_2]
})
print('n--- dtypes ---')
print(df.dtypes)
print('n--- selected: dt == dt ---')
selected = df[ df['ds'] == tomorrow_1 ]
print(selected)
print('n--- selected: dt == dt.strftime() ---')
selected = df[ df['ds'] == tomorrow_1.strftime('%Y-%m-%d') ]
print(selected)
print('n--- selected: dt.strftime() == dt.strftime() ---')
selected = df[ df['ds'].dt.strftime('%Y-%m-%d') == tomorrow_1.strftime('%Y-%m-%d') ]
print(selected)
print('n--- selected: dt.date() == dt.date() ---')
selected = df[ df['ds'].dt.date == tomorrow_1.date() ]
print(selected)

结果：

tomorrow_1: 2021-07-11 17:31:50.259787 | strftime: 2021-07-11
tomorrow_2: 2021-07-11 17:31:52.260139 | strftime: 2021-07-11
dt == dt            : False
dt == dt.strftime() : False
dt.strftime() == dt.strftime() : True
dt.date() == dt.date()     : True
--- dtypes ---
ds    datetime64[ns]
dtype: object
--- selected: dt == dt ---
ds
0 2021-07-11 17:31:50.259787
--- selected: dt == dt.strftime() ---
Empty DataFrame
Columns: [ds]
Index: []
--- selected: dt.strftime() == dt.strftime() ---
ds
0 2021-07-11 17:31:50.259787
1 2021-07-11 17:31:52.260139
--- selected: dt.date() == dt.date() ---
ds
0 2021-07-11 17:31:50.259787
1 2021-07-11 17:31:52.260139

---

编辑：

其他示例

import datetime
one_day = datetime.timedelta(days=1)
today     = datetime.datetime.today()
yesterday = today - one_day
tomorrow  = today + one_day
two_days_later = today + 2*one_day
# --- generate some data ---
import pandas as pd
forecast = pd.DataFrame({
'ds': [yesterday, today, tomorrow, two_days_later],
'yhat': [111, 222, 333, 444],
})
print('- forecast -')
print(forecast)
# --- find tomorrow ---
tomorrow = datetime.datetime.today() + datetime.timedelta(days=1)
mask = ( forecast['ds'].dt.date == tomorrow.date() )
print('- mask -')
print(mask)
selected = forecast[ mask ]
print('- selected -')
print(selected['yhat'].item())