How can I concatenate the rows in a pyspark dataframe with multiple columns using groupby and aggregate

问题

I have a pyspark dataframe with multiple columns. For example the one below.

from pyspark.sql import Row
l = [('Jack',"a","p"),('Jack',"b","q"),('Bell',"c","r"),('Bell',"d","s")]
rdd = sc.parallelize(l)
score_rdd = rdd.map(lambda x: Row(name=x[0], letters1=x[1], letters2=x[2]))
score_card = sqlContext.createDataFrame(score_rdd)

+----+--------+--------+
|name|letters1|letters2|
+----+--------+--------+
|Jack|       a|       p|
|Jack|       b|       q|
|Bell|       c|       r|
|Bell|       d|       s|
+----+--------+--------+

Now I want to group by "name" and concatenate the values in every row for both columns. I know how to do it but let's say there are thousands of rows then my code becomes very ugly. Here is my solution.

import pyspark.sql.functions as f
t = score_card.groupby("name").agg(
    f.concat_ws("",collect_list("letters1").alias("letters1")),
    f.concat_ws("",collect_list("letters2").alias("letters2"))
)

Here is the output I get when I save it in a CSV file.

+----+--------+--------+
|name|letters1|letters2|
+----+--------+--------+
|Jack|      ab|      pq|
|Bell|      cd|      rs|
+----+--------+--------+

But my main concern is about these two lines of code

f.concat_ws("",collect_list("letters1").alias("letters1")),
f.concat_ws("",collect_list("letters2").alias("letters2"))

If there are thousands of columns then I will have to repeat the above code thousands of times. Is there a simpler solution for this so that I don't have to repeat f.concat_ws() for every column?

I have searched everywhere and haven't been able to find a solution.

回答1:

yes, you can use for loop inside agg function and iterate through df.columns. Let me know if it helps.

    from pyspark.sql import functions as F
    df.show()

    # +--------+--------+----+
    # |letters1|letters2|name|
    # +--------+--------+----+
    # |       a|       p|Jack|
    # |       b|       q|Jack|
    # |       c|       r|Bell|
    # |       d|       s|Bell|
    # +--------+--------+----+

    df.groupBy("name").agg( *[F.array_join(F.collect_list(column), "").alias(column) for column in df.columns if column !='name' ]).show()

    # +----+--------+--------+
    # |name|letters1|letters2|
    # +----+--------+--------+
    # |Bell|      cd|      rs|
    # |Jack|      ab|      pq|
    # +----+--------+--------+

来源：https://stackoverflow.com/questions/62784146/how-can-i-concatenate-the-rows-in-a-pyspark-dataframe-with-multiple-columns-usin