Last active
July 3, 2021 04:54
-
-
Save zhu3pang/91990e27fd3149d7dabd1a8519ab242f to your computer and use it in GitHub Desktop.
spark sql 列转行
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
df = spark.createDataFrame([ | |
{'id': 1, u'姓名': u'张三', u'分数': 88, u'科目': u'数学'}, | |
{'id': 2, u'姓名': u'李雷', u'分数': 67, u'科目': u'数学'}, | |
{'id': 3, u'姓名': u'宫九', u'分数': 77, u'科目': u'数学'}, | |
{'id': 4, u'姓名': u'王五', u'分数': 65, u'科目': u'数学'}, | |
{'id': 1, u'姓名': u'张三', u'分数': 77, u'科目': u'英语'}, | |
{'id': 2, u'姓名': u'李雷', u'分数': 90, u'科目': u'英语'}, | |
{'id': 3, u'姓名': u'宫九', u'分数': 24, u'科目': u'英语'}, | |
{'id': 4, u'姓名': u'王五', u'分数': 90, u'科目': u'英语'}, | |
{'id': 1, u'姓名': u'张三', u'分数': 33, u'科目': u'语文'}, | |
{'id': 2, u'姓名': u'李雷', u'分数': 87, u'科目': u'语文'}, | |
{'id': 3, u'姓名': u'宫九', u'分数': 92, u'科目': u'语文'}, | |
{'id': 4, u'姓名': u'王五', u'分数': 87, u'科目': u'语文'}, | |
]) | |
df.show() | |
# +---+----+----+----+ | |
# | id|分数|姓名|科目| | |
# +---+----+----+----+ | |
# | 1| 88|张三|数学| | |
# | 2| 67|李雷|数学| | |
# | 3| 77|宫九|数学| | |
# | 4| 65|王五|数学| | |
# | 1| 77|张三|英语| | |
# | 2| 90|李雷|英语| | |
# | 3| 24|宫九|英语| | |
# | 4| 90|王五|英语| | |
# | 1| 33|张三|语文| | |
# | 2| 87|李雷|语文| | |
# | 3| 92|宫九|语文| | |
# | 4| 87|王五|语文| | |
# +---+----+----+----+ | |
# spark sql 列转行 | |
spark_df_pivot = spark.sql(""" | |
SELECT | |
* | |
FROM df | |
PIVOT | |
( | |
SUM(`分数`) | |
FOR `科目` in ('数学','英语') | |
) | |
ORDER BY id | |
""").cache() | |
spark_df_pivot.show() | |
# +---+----+----+----+ | |
# | id|姓名|数学|英语| | |
# +---+----+----+----+ | |
# | 1|张三| 88| 77| | |
# | 2|李雷| 67| 90| | |
# | 3|宫九| 77| 24| | |
# | 4|王五| 65| 90| | |
# +---+----+----+----+ | |
# spark scala | |
val spark_scala_pivot = | |
df.groupBy("id", "姓名") | |
.pivot("科目") | |
.agg(sum("分数")) | |
spark_scala_pivot.show() | |
# +---+----+----+----+----+ | |
# | id|姓名|数学|英语|语文| | |
# +---+----+----+----+----+ | |
# | 2|李雷| 67| 90| 87| | |
# | 1|张三| 88| 77| 33| | |
# | 3|宫九| 77| 24| 92| | |
# | 4|王五| 65| 90| 87| | |
# +---+----+----+----+----+ | |
# hive sql 列转行 | |
hive_df_pivot = spark.sql(""" | |
select | |
id, | |
`姓名`, | |
nvl(m ['数学'], 0) as `数学`, | |
nvl(m ['英语'], 0) as `英语` | |
from( | |
select | |
id, | |
`姓名`, | |
collect(`科目`, `分数`) as m | |
from | |
df | |
group by | |
id, | |
`姓名` | |
) t_b | |
""").cache() | |
hive_df_pivot.show() | |
# id 姓名 数学 英语 | |
# 4 王五 65 90 | |
# 1 张三 88 77 | |
# 2 李雷 67 90 | |
# 3 宫九 77 24 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
see how-to-pivot-spark-dataframe
and reshaping-pivoting-data-in-spark-rdd-and-or-spark-dataframes
and sql-ref-syntax-qry-select-pivot