Python Pandas：日期索引和下一个日期索引之间的日期时间项计数

Question

Python 3.3.3 熊猫 0.12.0

总的来说，我对 Pandas 和 Python 比较陌生。我有一个使用日期时间索引的 DataFrame（随时间推移的工作时间）。我还有一个我休假的日期列表（也是日期时间格式）。

我已将 DataFrame 重新采样为 28 天的区块，我希望添加一个列，其中包含在该 28 天的区块中休假的天数。

我确信有一些很棒的方法可以通过列表理解来做到这一点，但我能想到的最好方法是使用几个 for 循环来生成一个列表，然后在 DataFrame 中使用该列表。

我希望有人能指点我以一种不那么笨重的方式来实现这一点。提前感谢您的任何建议。

num_vaca = []

for idx, month_start in enumerate(df.index):
    counter = 0
    for day in vaca_days:
        if day >= month_start and day <= df.index[idx + 1]:
            counter += 1
    num_vaca.append(counter)

df['Vacation Days'] = num_vaca

这似乎行得通好的，我只是确定有更好的方法。

编辑：我没有很好地解释这个问题，我认为一些例子可能会有所帮助。

vaca_days = [datetime.datetime(2013, 1, 1, 0, 0),
 datetime.datetime(2013, 1, 2, 0, 0),
 datetime.datetime(2013, 1, 3, 0, 0),
 datetime.datetime(2013, 2, 4, 0, 0),
 datetime.datetime(2013, 2, 5, 0, 0),
 datetime.datetime(2013, 2, 6, 0, 0),
 datetime.datetime(2013, 1, 7, 0, 0),
 datetime.datetime(2013, 1, 8, 0, 0),
 datetime.datetime(2013, 1, 9, 0, 0)]

dr = pd.date_range('11/1/2012', periods = 6, freq = 'M')
df = pd.DataFrame(np.random.randn(len(dr)), index = dr, columns = ['Random'])
df

# Output:
              Random
2012-11-30 -0.134878
2012-12-31  0.066865
2013-01-31  0.630906
2013-02-28 -0.512881
2013-03-31  0.361890
2013-04-30 -1.019923


# Code from above
num_vaca = []

for idx, month_start in enumerate(df.index):
    counter = 0
    for day in vaca_days:
        if day >= month_start and day <= df.index[idx + 1]:
            counter += 1
    num_vaca.append(counter)

df['Vacation Days'] = num_vaca

df
# Output
              Random  Vacation Days
2012-11-30 -0.134878              0
2012-12-31  0.066865              6
2013-01-31  0.630906              3
2013-02-28 -0.512881              0
2013-03-31  0.361890              0
2013-04-30 -1.019923              0

在处理最后一个索引值和[idx + 1] 时仍然存在一些问题，因为我还没有定义该时间段应该在哪里停止。

---

2014 年 2 月 10 日更新

我能够得到@Andy-Hayden 对工作的回应，只是有点肘部油脂。以下是我遇到的问题，以及我能找到的最佳解决方法。

#! /usr/bin/env python3

import pandas as pd
import numpy as np
import datetime

# Generate a DataFrame with some random data and 28d date periods
dr = pd.date_range('11/6/2012', periods = 12, freq = 'W-Tue')
df = pd.DataFrame(np.random.randn(len(dr)), index = dr, columns = ['column1'])
df = df.resample(rule = '28D', how = 'mean')

# The start dates for each period
df
#              column1
# 2012-11-06 -0.134757
# 2012-12-04 -0.382133
# 2013-01-01 -0.468343
# 2013-01-29  0.000000
# [4 rows x 1 columns]

# Vacation dates
vaca_days = list(pd.date_range(start = '2012-12-1', periods = 12, freq = 'D'))
vaca_days
# Timestamp('2012-12-01 00:00:00', tz=None),
#  Timestamp('2012-12-02 00:00:00', tz=None),
#  Timestamp('2012-12-03 00:00:00', tz=None),
#  Timestamp('2012-12-04 00:00:00', tz=None),
#  Timestamp('2012-12-05 00:00:00', tz=None),
#  Timestamp('2012-12-06 00:00:00', tz=None),
#  Timestamp('2012-12-07 00:00:00', tz=None),
#  Timestamp('2012-12-08 00:00:00', tz=None),
#  Timestamp('2012-12-09 00:00:00', tz=None),
#  Timestamp('2012-12-10 00:00:00', tz=None),
#  Timestamp('2012-12-11 00:00:00', tz=None),
#  Timestamp('2012-12-12 00:00:00', tz=None)]

# Clearly there should be 3 vacation dates in the 2012-11-06 period
# and 9 vacation dates in the 2013-12-04 period
# ---------------------------
# This gives me wrong numbers, because it tries to start the period on vaca_days[0]
pd.Series(1, vaca_days).resample('28D', how='sum')
# 2012-12-01    12
# 2012-12-29     0
# Freq: 28D, dtype: int64

# No help here
pd.Series(1, vaca_days).resample('28D', how='sum').shift(-1)
# 2012-12-01     0
# 2012-12-29   NaN
# Freq: 28D, dtype: float64

# The only workaround I could figure out: Append on the first start day
vaca_days.append(pd.Timestamp('2012-11-06'))

vaca_series = pd.Series([0] + [1 for each in vaca_days[:-1]], sorted(vaca_days))
vaca_series = vaca_series.resample(rule = '28D', how = sum)

vaca_series
# 2012-11-06    3
# 2012-12-04    9
# 2013-01-01    0
# Freq: 28D, dtype: int64

df['Vacation Days'] = vaca_series
df['Vacation Days'] = df['Vacation Days'].fillna(0)
df

#                   column1  Vacation Days
# 2012-11-06  -4.761727e-01              3
# 2012-12-04   5.715697e-01              9
# 2013-01-01   4.065451e-02              0
# 2013-01-29  2.781364e-309              0
# [4 rows x 2 columns]

Answer 1

您可以使用 sum 作为聚合函数按月resample：

In [11]: pd.Series(1, vaca_days).resample('M', how='sum')
Out[11]: 
2013-01-31    6
2013-02-28    3
Freq: M, dtype: int64

当您将其设置为列时，它会为您对齐索引：

In [12]: df['vac_days'] = pd.Series(1, vaca_days).resample('M', how='sum')

In [13]: df
Out[13]: 
              Random  vac_days
2012-11-30 -0.134878       NaN
2012-12-31  0.066865       NaN
2013-01-31  0.630906         6
2013-02-28 -0.512881         3
2013-03-31  0.361890       NaN
2013-04-30 -1.019923       NaN

[6 rows x 2 columns]

现在您肯定希望fill in the NaN 使用 0。鉴于您想要的解决方案，您似乎也想将结果增加一 shift...

In [14]: df['vac_days'].shift(-1).fillna(0)
Out[14]: 
2012-11-30    0
2012-12-31    6
2013-01-31    3
2013-02-28    0
2013-03-31    0
2013-04-30    0
Name: vac_days, dtype: float64

In [15]: df['vac_days'] = df['vac_days'].shift(-1).fillna(0)