Sparklyr:使用group_by,然后从组中的行连接字符串

Question

我试图在sparklyr中使用group_by()和mutate()函数来连接组中的行.

这是一个我认为应该有效的简单示例,但不是:

library(sparkylr)
d <- data.frame(id=c("1", "1", "2", "2", "1", "2"), 
             x=c("200", "200", "200", "201", "201", "201"), 
             y=c("This", "That", "The", "Other", "End", "End"))
d_sdf <- copy_to(sc, d, "d")
d_sdf %>% group_by(id, x) %>% mutate( y = paste(y, collapse = " "))

我想要它产生的是:

Source: local data frame [6 x 3]
Groups: id, x [4]

# A tibble: 6 x 3
      id      x         y
  <fctr> <fctr>     <chr>
1      1    200 This That
2      1    200 This That
3      2    200       The
4      2    201 Other End
5      1    201       End
6      2    201 Other End

我收到以下错误:

Error: org.apache.spark.sql.AnalysisException: missing ) at 'AS' near '' '' in selection target; line 1 pos 42

请注意,在data.frame上使用相同的代码可以正常工作:

d %>% group_by(id, x) %>% mutate( y = paste(y, collapse = " "))

Answer 1

Spark sql如果您在没有聚合的情况下使用聚合函数,那么它就不喜欢它了,因此这个函数在dplyr普通函数中工作dataframe但不在函数中SparkDataFrame- sparklyr将命令转换为sql语句.如果您查看错误消息中的第二位,您可以观察到这种错误:

== SQL ==
SELECT `id`, `x`, CONCAT_WS(' ', `y`, ' ' AS "collapse") AS `y`

paste被翻译成CONCAT_WS.concat但是会将列粘贴在一起.

一个更好的等价会collect_list和collect_set,但它们产生list的输出.

但你可以建立在:

如果你不希望有同一行中的结果复制就可以使用summarise,collect_list以及paste:

res <- d_sdf %>% 
      group_by(id, x) %>% 
      summarise( yconcat =paste(collect_list(y)))

结果:

Source:     lazy query [?? x 3]
Database:   spark connection master=local[8] app=sparklyr local=TRUE
Grouped by: id

     id     x         y
  <chr> <chr>     <chr>
1     1   201       End
2     2   201 Other End
3     1   200 This That
4     2   200       The

你可以加入这个返回到原来的数据,如果你不希望你的行复制:

d_sdf %>% left_join(res)

结果:

Source:     lazy query [?? x 4]
Database:   spark connection master=local[8] app=sparklyr local=TRUE

     id     x     y   yconcat
  <chr> <chr> <chr>     <chr>
1     1   200  This This That
2     1   200  That This That
3     2   200   The       The
4     2   201 Other Other End
5     1   201   End       End
6     2   201   End Other End