<p>你可以这样做:</p>
<pre><code>>>> df
>>>
fruit amount
20140101 apple 3
20140102 apple 5
20140102 orange 10
20140104 banana 2
20140104 apple 10
20140104 orange 4
20140105 orange 6
20140105 grape 1
>>> g= df.set_index('fruit', append=True).groupby(level=1)
>>> res = g['amount'].apply(pd.rolling_mean, 3, 1).reset_index('fruit')
>>> res
fruit 0
20140101 apple 3.000000
20140102 apple 4.000000
20140102 orange 10.000000
20140104 banana 2.000000
20140104 apple 6.000000
20140104 orange 7.000000
20140105 orange 6.666667
20140105 grape 1.000000
</code></pre>
<p><strong>更新</strong></p>
<p>好吧,正如@cphlewis在评论中提到的,我的代码不会给出您想要的结果。我已经检查了不同的方法,到目前为止我发现的方法是这样的(但不确定性能):</p>
<pre><code>>>> df.index = [pd.to_datetime(str(x), format='%Y%m%d') for x in df.index]
>>> df.reset_index(inplace=True)
>>> def avg_3_days(x):
return df[(df['index'] >= x['index'] - pd.DateOffset(3)) & (df['index'] < x['index']) & (df['fruit'] == x['fruit'])].amount.mean()
>>> df['res'] = df.apply(avg_3_days, axis=1)
>>> df
index fruit amount res
0 2014-01-01 apple 3 NaN
1 2014-01-02 apple 5 3
2 2014-01-02 orange 10 NaN
3 2014-01-04 banana 2 NaN
4 2014-01-04 apple 10 4
5 2014-01-04 orange 4 10
6 2014-01-05 orange 6 7
7 2014-01-05 grape 1 NaN
</code></pre>