python - 如何找到一个日期和另一个日期之间的差异(如果第二个未完全插入到dataFrame中)

Question

几天来我一直被这个问题困扰着……我不知道如何解决。
我在索引中有一些日期的数据框。我想选择一个等于天数的窗口，例如说5天。我想返回一个dataFrame，其索引中包含我的初始日期，并且在第一列中，索引中的每个日期与过去5天的最近天之间的天数差。

让我们举个例子。

[In] Mydates
[Out] 
2017-04-04   
2017-04-03    
2017-03-31    
2017-03-30   
2017-03-29   
2017-03-28   
2017-03-27   
2017-03-24  
2017-03-23     
2017-03-21   

我要回来

func(window = 5)
    return MyNewdates
[out]         First column
2017-04-04   -5 [diff between 2017-04-04 and 5 days before or closest date in dataset from 5 days before (here 2017-03-30 ), so difference is 0 - 5 =] -5 

2017-04-03   -5  [diff between 2017-04-03 and 5 days before or closest date in dataset from 5 days before (here 2017-03-29), so difference is 0 - 5 =] -5 
2017-03-31    
2017-03-30   -6  [here, there is no 2017-03-25 (5 days before) so the closest date from my window is 2017-03-24 (6 days before), so the difference is 0 - 6 =] -6 

2017-03-29   -5  [diff between 2017-03-29 and 5 days before or closest date in dataset from 5 days before (here 2017-03-24), so difference is 0 - 5 =] -5 
2017-03-28   -5  [diff between 2017-03-29 and 5 days before or closest date in dataset from 5 days before (here 2017-03-23 ), so difference is 0 - 5 =] -5 
2017-03-27   -4  [diff between 2017-03-27 and 5 days before or closest date in dataset from 5 days before (here 2017-03-23 ), so difference is 0 - 4 =] -4
2017-03-24  NAN 
2017-03-23  NAN    
2017-03-21  NAN

等等...

为此，我将所有日期转换为几天。还有其他方法吗？我希望它能使我回想起过去的时差。

希望一切都清楚，如果您有任何疑问，请告诉我！

谢谢！！

Answer 1

IIUC然后根据需要进行以下工作：

In [141]:
import io
import pandas as pd
# read in data
t="""Dates
2017-04-04   
2017-04-03    
2017-03-31    
2017-03-30   
2017-03-29   
2017-03-28   
2017-03-27   
2017-03-24  
2017-03-23   
2017-03-22   
2017-03-21  """
df = pd.read_csv(io.StringIO(t), delim_whitespace=True, parse_dates=[0], index_col=[0])
# define a window func
def func(x, window):
    prev = x - pd.DateOffset(window)
    if df.index.isin([prev]).any() == True:
        return -window
    elif (prev < df.index).all():
        return np.NaN
    else:
        diff = (df.index - prev).to_series().abs() 
        diff_idx = diff.index.get_loc(diff.argmin())
        return - ((x - (x - df.iloc[diff_idx]).name).days)

df.index.to_series().apply(lambda x: func(x, 5))
Out[141]:

Dates
2017-04-04   -5.0
2017-04-03   -5.0
2017-03-31   -4.0
2017-03-30   -6.0
2017-03-29   -5.0
2017-03-28   -5.0
2017-03-27   -5.0
2017-03-24    NaN
2017-03-23    NaN
2017-03-22    NaN
2017-03-21    NaN
Name: Dates, dtype: float64