在一个dask数据帧中，一行与其前导之间的差异为3行

我有一个CSV文件，它通过以下代码导入为dask数据帧：

import dask.dataframe as dd
df = dd.read_csv("name and path of the file.csv")
df.head(10)
产量

+-----+------+-----+
| A | 2 | 4 |
+-----+------+-----+
| A | 4 | 5 |
+-----+------+-----+
| A | 7 | 7 |
+-----+------+-----+
| A | 3 | 8 |
+-----+------+-----+
| A | 7 | 3 |
+-----+------+-----+
| B | 8 | 9 |
+-----+------+-----+
| B | 10 | 10 |
+-----+------+-----+
| B | 8 | 9 |
+-----+------+-----+

我想创建另一列col4，分别包含col2[n+3]/col2-1col1 中的每个组。

输出应该是

+-----+------+-----+-----+
|col1 | col2 | col3| col4|
+-----+------+-----+-----+
| A | 2 | 4 | 0.5| #(3/2-1)
+-----+------+-----+-----+
| A | 4 | 5 | 0.75| #(7/4-1)
+-----+------+-----+-----+
| A | 7 | 7 | NA |
+-----+------+-----+-----+
| A | 3 | 8 | NA |
+-----+------+-----+-----+
| A | 7 | 3 | NA |
+-----+------+-----+-----+
| B | 8 | 9 | 1.5 |
+-----+------+-----+-----+
| B | 10 | 10 | NA |
+-----+------+-----+-----+
| B | 8 | 9 | NA |
+-----+------+-----+-----+
| B | 20 | 15 | NA |
+-----+------+-----+-----+
我们可以在熊猫上完成这项任务，如下所示

df['col4'] = df.groupby('col1')['col2'].transform(lambda x: x.shift(-3)) / df['col2'] - 1
但它在dask中不起作用。