以天爲單位獲得Pandas DatetimeIndex系列差異的最佳方法？

Question

假設我有一系列pandas.tseries.index.DatetimeIndex，它基本上是2016年的平日。是否有任何簡單/優雅的方式來查找系列中天數的順序差異？就像.diff（）爲整數或浮點型DataFrame列一樣。

import pandas as pd 
import numpy as np 
ds = pd.date_range("2016-01-01","2016-12-31",freq='B') 

# I was hoping for something like this: 
ds.diff().days 

# this gives me what I want, but it is ugly and unintuitive 
np.diff(ds)/86400000000000 

我也想過np.diff(ds.date)，但它給了我datetime.timedelta的ndarray，我不知道如何把它改造成一個整數數組/系列沒有一個for循環。

Answer 1

試試這個：

In [154]: ds.to_series().diff() 
Out[154]: 
2016-01-01  NaT 
2016-01-04 3 days 
2016-01-05 1 days 
2016-01-06 1 days 
2016-01-07 1 days 
2016-01-08 1 days 
2016-01-11 3 days 
2016-01-12 1 days 
2016-01-13 1 days 
2016-01-14 1 days 
2016-01-15 1 days 
2016-01-18 3 days 
2016-01-19 1 days 
2016-01-20 1 days 
2016-01-21 1 days 
2016-01-22 1 days 
2016-01-25 3 days 
2016-01-26 1 days 
2016-01-27 1 days 
2016-01-28 1 days 
2016-01-29 1 days 
2016-02-01 3 days 
2016-02-02 1 days 
2016-02-03 1 days 
2016-02-04 1 days 
       ... 

Answer 2

我想你需要添加Index.to_series然後Series.dt.days如果需要數字輸出：

print (ds.to_series().diff().dt.days) 

2016-01-01 NaN 
2016-01-04 3.0 
2016-01-05 1.0 
2016-01-06 1.0 
2016-01-07 1.0 
2016-01-08 1.0 
2016-01-11 3.0 
2016-01-12 1.0 
2016-01-13 1.0 
2016-01-14 1.0 
2016-01-15 1.0 
2016-01-18 3.0 
2016-01-19 1.0 
2016-01-20 1.0 
2016-01-21 1.0 
2016-01-22 1.0 
2016-01-25 3.0 
2016-01-26 1.0 
2016-01-27 1.0 
2016-01-28 1.0 
2016-01-29 1.0 
2016-02-01 3.0 
2016-02-02 1.0 
2016-02-03 1.0 
2016-02-04 1.0 
2016-02-05 1.0 
2016-02-08 3.0 
2016-02-09 1.0 
2016-02-10 1.0 
2016-02-11 1.0 

2016-11-21 3.0 
2016-11-22 1.0 
2016-11-23 1.0 
2016-11-24 1.0 
2016-11-25 1.0 
2016-11-28 3.0 
2016-11-29 1.0 
2016-11-30 1.0 
2016-12-01 1.0 
2016-12-02 1.0 
2016-12-05 3.0 
2016-12-06 1.0 
2016-12-07 1.0 
2016-12-08 1.0 
2016-12-09 1.0 
2016-12-12 3.0 
2016-12-13 1.0 
2016-12-14 1.0 
2016-12-15 1.0 
2016-12-16 1.0 
2016-12-19 3.0 
2016-12-20 1.0 
2016-12-21 1.0 
2016-12-22 1.0 
2016-12-23 1.0 
2016-12-26 3.0 
2016-12-27 1.0 
2016-12-28 1.0 
2016-12-29 1.0 
2016-12-30 1.0 
Freq: B, dtype: float64