我找到了如何重新采样多索引的描述:
Resampling Within a Pandas MultiIndex
然而,只要我使用count而不是sum,解决方案就不再起作用了
这可能与:Resampling with 'how=count' causing problems
有关无效计数和字符串:
values_a =[1]*16
states = ['Georgia']*8 + ['Alabama']*8
#cities = ['Atlanta']*4 + ['Savanna']*4 + ['Mobile']*4 + ['Montgomery']*4
dates = pd.DatetimeIndex([datetime.datetime(2012,1,1)+datetime.timedelta(days = i) for i in range(4)]*4)
df2 = pd.DataFrame(
{'value_a': values_a},
index = [states, dates])
df2.index.names = ['State', 'Date']
df2.reset_index(level=[0], inplace=True)
print(df2.groupby(['State']).resample('W',how='count'))
收率:
2012-01-01 2012-01-08
State value_a State value_a
State
Alabama 2 2 6 6
Georgia 2 2 6 6
总和和数字作为值的工作版
values_a =[1]*16
states = ['Georgia']*8 + ['Alabama']*8
#cities = ['Atlanta']*4 + ['Savanna']*4 + ['Mobile']*4 + ['Montgomery']*4
dates = pd.DatetimeIndex([datetime.datetime(2012,1,1)+datetime.timedelta(days = i) for i in range(4)]*4)
df2 = pd.DataFrame(
{'value_a': values_a},
index = [states, dates])
df2.index.names = ['State', 'Date']
df2.reset_index(level=[0], inplace=True)
print(df2.groupby(['State']).resample('W',how='sum'))
收益率(注意“州”不重复):
value_a
State Date
Alabama 2012-01-01 2
2012-01-08 6
Georgia 2012-01-01 2
2012-01-08 6
答案 0 :(得分:1)
使用count
时,状态不是一个讨厌的列(它可以计算字符串),因此resample
将对它应用计数(尽管输出不是我所期望的)。你可以这样做(告诉它只适用count
到value_a
),
>>> print df2.groupby(['State']).resample('W',how={'value_a':'count'})
value_a
State Date
Alabama 2012-01-01 2
2012-01-08 6
Georgia 2012-01-01 2
2012-01-08 6
或者更一般地说,您可以将不同类型的how
应用于不同的列:
>>> print df2.groupby(['State']).resample('W',how={'value_a':'count','State':'last'})
State value_a
State Date
Alabama 2012-01-01 Alabama 2
2012-01-08 Alabama 6
Georgia 2012-01-01 Georgia 2
2012-01-08 Georgia 6
因此,虽然以上允许您count
重新采样的多索引数据帧,但它不能解释how='count'
的输出行为。以下内容更接近我期望它的表现方式:
print df2.groupby(['State']).resample('W',how={'value_a':'count','State':'count'})
State value_a
State Date
Alabama 2012-01-01 2 2
2012-01-08 6 6
Georgia 2012-01-01 2 2
2012-01-08 6 6
答案 1 :(得分:1)
@Karl D soln是正确的;这将在0.14 / master(即将发布)中实现,请参阅文档here
In [118]: df2.groupby([pd.Grouper(level='Date',freq='W'),'State']).count()
Out[118]:
value_a
Date State
2012-01-01 Alabama 2
Georgia 2
2012-01-08 Alabama 6
Georgia 6
在0.14之前,很难用基于时间的石斑鱼和另一个石斑鱼进行分组/重新采样。 pd.Grouper
允许非常灵活的规范来执行此操作。