Question

给定时间序列s，带有日期时间索引，我希望能够按日期字符串索引时间序列。我误解了这应该如何运作？

import pandas as pd
url = 'http://ichart.finance.yahoo.com/table.csvs=SPY&d=12&e=4&f=2012&g=d&a=01&b=01&c=2001&ignore=.csv'  
df = pd.read_csv(url, index_col='Date', parse_dates=True)  
s = df['Close']
s['2012-12-04']

结果：

TimeSeriesError                           Traceback (most recent call last)
<ipython-input-244-e2ccd4ecce94> in <module>()
      2 df = pd.read_csv(url, index_col='Date', parse_dates=True)  
      3 s = df['Close']  
----> 4 s['2012-12-04']

    G:\Python27-32\lib\site-packages\pandas\core\series.pyc in __getitem__(self, key)
    468     def __getitem__(self, key):
    469         try:
--> 470             return self.index.get_value(self, key)
    471         except InvalidIndexError:
    472             pass

    G:\Python27-32\lib\site-packages\pandas\tseries\index.pyc in get_value(self, series, key)  

   1030 
   1031             try:
-> 1032                 loc = self._get_string_slice(key)
   1033                 return series[loc]
   1034             except (TypeError, ValueError, KeyError):

G:\Python27-32\lib\site-packages\pandas\tseries\index.pyc in _get_string_slice(self, key)
   1077         asdt, parsed, reso = parse_time_string(key, freq)
   1078         key = asdt
-> 1079         loc = self._partial_date_slice(reso, parsed)
   1080         return loc
   1081 

G:\Python27-32\lib\site-packages\pandas\tseries\index.pyc in _partial_date_slice(self, reso, parsed)
    992     def _partial_date_slice(self, reso, parsed):
    993         if not self.is_monotonic:
--> 994             raise TimeSeriesError('Partial indexing only valid for ordered '
    995                                   'time series.')
    996 

TimeSeriesError: Partial indexing only valid for ordered time series.

更具体（也许是迂腐......），这里的2个时间序列有什么区别：

import pandas as pd
url = 'http://ichart.finance.yahoo.com/table.csv?        s=SPY&d=12&e=4&f=2012&g=d&a=01&b=01&c=2001&ignore=.csv'
s = pd.read_csv(url, index_col='Date', parse_dates=True)['Close']
rng = date_range(start='2011-01-01', end='2011-12-31')
ts = Series(randn(len(rng)), index=rng)
print ts.__class__
print ts.index[0].__class__
print s1.__class__
print s1.index[0].__class__
print ts[ts.index[0]]
print s[s.index[0]]
print ts['2011-01-01']
try:
    print s['2012-12-05']
except:
    print "doesn't work"

结果：

<class 'pandas.core.series.TimeSeries'>
<class 'pandas.lib.Timestamp'>
<class 'pandas.core.series.TimeSeries'>
<class 'pandas.lib.Timestamp'>
-0.608673793503
141.5
-0.608673793503
doesn't work

Answer 1

尝试使用Timestamp对象建立索引：

>>> import pandas as pd
>>> from pandas.lib import Timestamp
>>> url = 'http://ichart.finance.yahoo.com/table.csv?s=SPY&d=12&e=4&f=2012&g=d&a=01&b=01&c=2001&ignore=.csv'
>>> df = pd.read_csv(url, index_col='Date', parse_dates=True)
>>> s = df['Close']
>>> s[Timestamp('2012-12-04')]
141.25

Answer 2

如果时间序列未订购并且您提供了部分时间戳（例如日期，而不是日期时间），则不清楚应选择哪个日期时间。

不能假设每个日期只有一个日期时间对象，尽管在这个例子中，这里有几个选项但是在这里抛出错误而不是猜测用户的动机似乎更安全。（我们可以返回一个类似于.ix['2011-01']的系列/列表，但如果在其他情况下返回一个数字，这可能会让人感到困惑。我们可以尝试返回“最接近的匹配”......但这并不是很有意义无论是。）

在有序的情况下，我们会选择第一个具有所选日期的日期时间。

您可以在此简单示例中看到此行为：

import pandas as pd
from numpy.random import randn
from random import shuffle
rng = pd.date_range(start='2011-01-01', end='2011-12-31')
rng2 = list(rng)
shuffle(rng2) # not in order
rng3 = list(rng)
del rng3[20] # in order, but no freq

ts = pd.Series(randn(len(rng)), index=rng)
ts2 = pd.Series(randn(len(rng)), index=rng2)
ts3 = pd.Series(randn(len(rng)-1), index=rng3)

ts.index
<class 'pandas.tseries.index.DatetimeIndex'>
[2011-01-01 00:00:00, ..., 2011-12-31 00:00:00]
Length: 365, Freq: D, Timezone: None

ts['2011-01-01']
# -1.1454418070543406

ts2.index
<class 'pandas.tseries.index.DatetimeIndex'>
[2011-04-16 00:00:00, ..., 2011-03-10 00:00:00]
Length: 365, Freq: None, Timezone: None

ts2['2011-01-01']
#...error which you describe
TimeSeriesError: Partial indexing only valid for ordered time series

ts3.index
<class 'pandas.tseries.index.DatetimeIndex'>
[2011-01-01 00:00:00, ..., 2011-12-31 00:00:00]
Length: 364, Freq: None, Timezone: None

ts3['2011-01-01']
1.7631554507355987


rng4 = pd.date_range(start='2011-01-01', end='2011-01-31', freq='H')
ts4 = pd.Series(randn(len(rng4)), index=rng4)

ts4['2011-01-01'] == ts4[0]
# True # it picks the first element with that date

我不认为这是一个错误，但我将其发布为an issue on github。

Answer 3

虽然大熊猫教程很有启发性，但我认为提出的原始问题值得直接回答。我遇到了将Yahoo图表信息转换为可以切片等的DataFrame的同样问题。我发现唯一需要的是：

import pandas as pd
import datetime as dt

def dt_parser(date): 
return dt.datetime.strptime(date, '%Y-%m-%d') + dt.timedelta(hours=16)

url = 'http://ichart.finance.yahoo.com/table.csvs=SPY&d=12&e=4&f=2012&g=d&a=01&b=01&c=2001&ignore=.csv'  
df = pd.read_csv(url, index_col=0, parse_dates=True, date_parser=dt_parser)
df.sort_index(inplace=True)
s = df['Close']
s['2012-12-04']     # now should work

“技巧”包括我自己的date_parser。我猜测在read_csv中有更好的方法可以做到这一点，但这至少产生了一个被索引并可以被切片的DataFrame。

按日期字符串索引时间序列

3 个答案: