python - PySpark - 将列表的列转换为行

Question

我有一个 pyspark 数据框。我必须做一个分组，然后将某些列聚合到一个列表中，这样我就可以在数据框上应用 UDF。

例如，我创建了一个数据框，然后按人分组。

df = spark.createDataFrame(a, ["Person", "Amount","Budget", "Date"])
df = df.groupby("Person").agg(F.collect_list(F.struct("Amount", "Budget", "Date")).alias("data"))
df.show(truncate=False)
+------+----------------------------------------------------------------------------+
|Person|data                                                                        |
+------+----------------------------------------------------------------------------+
|Bob   |[[85.8,Food,2017-09-13], [7.8,Household,2017-09-13], [6.52,Food,2017-06-13]]|
+------+----------------------------------------------------------------------------+ 

我省略了 UDF，但 UDF 生成的数据框在下方。

+------+--------------------------------------------------------------+
|Person|res                                                           |
+------+--------------------------------------------------------------+
|Bob   |[[562,Food,June,1], [380,Household,Sept,4], [880,Food,Sept,2]]|
+------+--------------------------------------------------------------+

我需要将生成的数据框转换为行，其中列表中的每个元素都是一个新行和一个新列。这可以在下面看到。

+------+------------------------------+
|Person|Amount|Budget   |Month|Cluster|
+------+------------------------------+
|Bob   |562   |Food     |June |1      |
|Bob   |380   |Household|Sept |4      |
|Bob   |880   |Food     |Sept |2      |
+------+------------------------------+

Answer 1

您可以使用 explode 和 getItem 如下:

# starting from this form:
+------+--------------------------------------------------------------
|Person|res                                                          |
+------+--------------------------------------------------------------+
|Bob   |[[562,Food,June,1], [380,Household,Sept,4], [880,Food,Sept,2]]|
+------+--------------------------------------------------------------+
import pyspark.sql.functions as F

# explode res to have one row for each item in res
exploded_df = df.select("*", F.explode("res").alias("exploded_data"))
exploded_df.show(truncate=False)

# then use getItem to create separate columns
exploded_df = exploded_df.withColumn(
            "Amount",
            F.col("exploded_data").getItem("Amount") # either get by name or by index e.g. getItem(0) etc
        )

exploded_df = exploded_df.withColumn(
            "Budget",
            F.col("exploded_data").getItem("Budget")
        )

exploded_df = exploded_df.withColumn(
            "Month",
            F.col("exploded_data").getItem("Month")
        )

exploded_df = exploded_df.withColumn(
            "Cluster",
            F.col("exploded_data").getItem("Cluster")
        )

exploded_df.select("Person", "Amount", "Budget", "Month", "Cluster").show(10, False)

+------+------------------------------+
|Person|Amount|Budget   |Month|Cluster|
+------+------------------------------+
|Bob   |562   |Food     |June |1      |
|Bob   |380   |Household|Sept |4      |
|Bob   |880   |Food     |Sept |2      |
+------+------------------------------+

然后您可以删除不需要的列。希望这对您有所帮助，祝您好运!