计算pandas中每组的数字差异

Question

我的Dataframe具有以下结构：

patient_id  |  timestamp  |  measurement
A           |  2014-10-10 |  5.7
A           |  2014-10-11 |  6.3
B           |  2014-10-11 |  6.1
B           |  2014-10-10 |  4.1

我想计算每位患者每次测量之间的delta（差异）。

结果如下：

patient_id  |  timestamp  |  measurement  |    delta
A           |  2014-10-10 |  5.7          |     NaN
A           |  2014-10-11 |  6.3          |     0.6
B           |  2014-10-11 |  6.1          |     2.0
B           |  2014-10-10 |  4.1          |     NaN

如何在熊猫中最优雅地完成这项工作？

Answer 1

在'measurement'列上调用transform并传递方法diff，transform会返回一个索引与原始df对齐的系列：

In [4]:

df['delta'] = df.groupby('patient_id')['measurement'].transform(pd.Series.diff)
df
Out[4]:
  patient_id   timestamp  measurement  delta
0          A  2014-10-10          5.7    NaN
1          A  2014-10-11          6.3    0.6
2          B  2014-10-10          4.1    NaN
3          B  2014-10-11          6.1    2.0

修改

如果您打算对transform的结果应用某些排序，请先对df进行排序：

In [10]:

df['delta'] = df.sort(columns=['patient_id', 'timestamp']).groupby('patient_id')['measurement'].transform(pd.Series.diff)
df
Out[10]:
  patient_id   timestamp  measurement  delta
0          A  2014-10-10          5.7    NaN
1          A  2014-10-11          6.3    0.6
2          B  2014-10-11          6.1    2.0
3          B  2014-10-10          4.1    NaN