Create monthly time series with arbitrary start dates

问题

Using pandas it is easy to create a monthly series of dates.

import pandas as pd

pd.date_range('2012-04-23', '2013-01-23', freq='BM')

DatetimeIndex(['2012-04-30', '2012-05-31', '2012-06-29', '2012-07-31',
               '2012-08-31', '2012-09-28', '2012-10-31', '2012-11-30',
               '2012-12-31'],
              dtype='datetime64[ns]', freq='BM')

Notice that the dates in the DatetimeIndex are month ends. I know that it should be considering I chose freq='BM' but I don't believe I had a choice that would have accomplished my goal.

I'm often in need for producing a monthly series of dates starting from the last business day going back in time every month.

I'd like to see this instead:

DatetimeIndex(['2012-04-23', '2012-05-23', '2012-06-23', '2012-07-23',
               '2012-08-23', '2012-09-23', '2012-10-23', '2012-11-23',
               '2012-12-23'],
              dtype='datetime64[ns]', freq=None)

or another more complicated example might be to get months from '2012-01-30' to '2012-04-30'. I'd expect to see:

DatetimeIndex(['2012-01-30', '2012-02-29', '2012-03-30', '2012-04-30'],
              dtype='datetime64[ns]', freq=None)

回答1:

You may be looking for something like this:

from pandas.tseries.offsets import Day, BDay
pd.date_range(start = '2012-01-01', periods = 6, freq = 'MS') + Day(22) + BDay(0)
Out[12]: 
DatetimeIndex(['2012-01-23', '2012-02-23', '2012-03-23', '2012-04-23',
               '2012-05-23', '2012-06-25'],
              dtype='datetime64[ns]', freq=None)

Day(22) adds an offset of 22 days and BDay is responsible for business day offset (BDay(0) takes the nearest business day).

It's a bit more difficult with dates starting at 30th. So I had to write a function for this. (For clarity of code it doesn't allow a custom freq parameter.)

def my_business_date_range(day, **kwargs):
    assert(isinstance(day, int) & (day > 0) & (day < 32))
    rng0 = pd.date_range(freq = 'MS', **kwargs)
    rng1 = rng0 + pd.tseries.offsets.Day(day-1) + pd.tseries.offsets.BDay(0)
    # Correcting overflows:
    overflow_idx, = np.nonzero(rng0.month != rng1.month)
    if overflow_idx.size > 0:
        # rng1 is not mutable
        tmp = rng1.tolist()        
        bme = pd.tseries.offsets.BusinessMonthEnd(-1)
        for i in overflow_idx:
            tmp[i] = bme(rng1[i])
        rng1 = pd.DatetimeIndex(tmp)
    return rng1

my_business_date_range(30, start= '2012-01-01', periods = 6)
Out[13]: 
DatetimeIndex(['2012-01-30', '2012-02-29', '2012-03-30', '2012-04-30',
               '2012-05-30', '2012-06-29'],
              dtype='datetime64[ns]', freq=None)

Pandas has also an experimental CustomBusinessMonth and the like but I couldn't make it work.

回答2:

I'm not clear on your question, but believe this is a move in the right direction.

start = '2012-04-23'
end = '2013-01-23'

>>> pd.DatetimeIndex([pd.datetime(ts.year, ts.month, int(end.split("-")[-1])) 
                      for ts in pd.date_range(start, end, freq='BM')])

DatetimeIndex(['2012-04-23', '2012-05-23', '2012-06-23', '2012-07-23', '2012-08-23', '2012-09-23', '2012-10-23', '2012-11-23', '2012-12-23'], dtype='datetime64[ns]', freq=None)

Although not optimized for speed, I believe the following function will return the correct values per your requirements.

def foo(date, periods, forward=True):
    if isinstance(date, str):
        date = pd.Timestamp(date).date()
    dates = [date + relativedelta(date, months=n * (1 if forward else -1)) for n in range(1, periods +1)]
    result = []
    print dates
    for date in dates:
        month = date.month
        iso_day = date.isoweekday()
        if iso_day == 6:
            date += dt.timedelta(days=2 if forward else -1)
        elif iso_day == 7:
            date += dt.timedelta(days=1 if forward else -2)
        if date.month != month:
            # Gone into next/preceding month.  Roll back/forward.
            date -= dt.timedelta(days=3 if forward else -3)
        result.append(date)
    return result

来源：https://stackoverflow.com/questions/37084925/create-monthly-time-series-with-arbitrary-start-dates

标签

python

pandas

python-datetime