所以我目前有一个数据帧，看起来如下：
+-------------+----------------+---------------+------------------+-----------------+
| customer_id | init_base_date | init_end_date | reinit_base_date | reinit_end_date |
+-------------+----------------+---------------+------------------+-----------------+
| ... | | | | |
| A | 2015-07-30 | | | |
| A | | 2016-07-24 | | |
| B | 2015-07-10 | | | |
| B | | 2015-10-05 | | |
| B | | | 2016-01-09 | |
| B | | | | 2016-07-04 |
| C | 2015-05-13 | | | |
| C | | 2015-08-09 | | |
| ... | | | | |
+-------------+----------------+---------------+------------------+-----------------+
我真的需要把它转换成
+-------------+----------------+---------------+------------------+-----------------+
| customer_id | init_base_date | init_end_date | reinit_base_date | reinit_end_date |
+-------------+----------------+---------------+------------------+-----------------+
| ... | | | | |
| A | 2015-07-30 | 2016-07-24 | | |
| B | 2015-07-10 | 2015-10-05 | 2016-01-09 | 2016-07-04 |
| C | 2015-05-13 | 2015-08-09 | | |
| ... | | | | |
+-------------+----------------+---------------+------------------+-----------------+
我能想出几种非常乏味的方法来完成上面的工作，但我想知道是否有一种快速有效的方法(也许是使用windows？我现在只使用PySpark一个月了，所以肯定还是个新手(。
如果显示的那些空单元格实际上是nulls(而不是空字符串(，则可以在groupBy中使用pyspark.sql.functions.first()作为聚合函数。关键是将first()的ignorenulls参数设置为True(默认为False(。
import pyspark.sql.functions as f
cols = [c for c in df.columns if c != 'customer_id']
df.groupBy('customer_id').agg(*[f.first(c, True).alias(c) for c in cols]).show()
#+-----------+--------------+-------------+----------------+---------------+
#|customer_id|init_base_date|init_end_date|reinit_base_date|reinit_end_date|
#+-----------+--------------+-------------+----------------+---------------+
#| A| 2015-07-30| 2016-07-24| null| null|
#| B| 2015-07-10| 2015-10-05| 2016-01-09| 2016-07-04|
#| C| 2015-05-13| 2015-08-09| null| null|
#+-----------+--------------+-------------+----------------+---------------+
如果这些空值实际上是空字符串，则可以先用null替换所有空字符串，然后按照上面的方法操作。然后可以(可选(将null值替换为空格。
from functools import reduce # for python3
cols = [c for c in df.columns if c != 'customer_id']
df = reduce(lambda df, c: df.withColumn(c, f.when(f.col(c) != '', f.col(c))), cols, df)
df = df.groupBy('customer_id').agg(*[f.first(c, True).alias(c) for c in cols])
df.na.fill('').show() # fill nulls with blanks
#+-----------+--------------+-------------+----------------+---------------+
#|customer_id|init_base_date|init_end_date|reinit_base_date|reinit_end_date|
#+-----------+--------------+-------------+----------------+---------------+
#| A| 2015-07-30| 2016-07-24| | |
#| B| 2015-07-10| 2015-10-05| 2016-01-09| 2016-07-04|
#| C| 2015-05-13| 2015-08-09| | |
#+-----------+--------------+-------------+----------------+---------------+