假设我们在工作目录中托管以下数据:
>library(sparklyr)
>library(dplyr)
>f<-data.frame(category=c("e","EE","W","S","Q","e","Q","S"),
DD=c(33.2,33.2,14.55,12,13.4,45,7,3),
CC=c(2,44,4,44,9,2,2.2,4),
>FF=c("A","A","A","A","A","A","B","A") )
>write.csv(f,"D.csv")##Write in working directory
我们使用spark命令从工作目录中读取文件
>sc <- spark_connect(master = "local", spark_home = "/home/tomas/spark-2.1.0-bin-hadoop2.7/", version = "2.1.0")
>df <- spark_read_csv(sc, name = "data", path = "D.csv", header = TRUE, delimiter = ",")
我想获得一个如下矩阵,其中按“类别”分组,求和DD,计算“ CC”的平均值,在“ FF”中计数不同
它会一直这样:
category SumDD MeanCC CountDistinctFF
e 78.2 2 1
EE 33.2 44. 1
WW 14.55 4 1
S 15 24 2
Q 20.4 5.6 1
答案 0 :(得分:1)
为了操纵spark DF,您需要使用dplyr函数。在星火环境中,Naveen的答案会起作用,除了最后一个变量。可以从dplyr尝试unique
来代替n_distinct
df0=df%>%group_by(category)%>%
summarize(sumDD=sum(DD,na.rm=T),MeanCC=mean(CC,na.rm=T),CountDistinctFF=n_distinct(FF))
要使用Spark DF检查结果,可以使用:
> glimpse(df0)
Observations: ??
Variables: 4
$ category <chr> "e", "EE", "S", "Q", "W"
$ sumDD <dbl> 78.20, 33.20, 15.00, 20.40, 14.55
$ MeanCC <dbl> 2.0, 44.0, 24.0, 5.6, 4.0
$ CountDistinctFF <dbl> 1, 1, 1, 2, 1
或者您可以将其收集回本地系统并像任何R数据框一样进行操作
> df0%>%collect
# A tibble: 5 x 4
category sumDD MeanCC CountDistinctFF
<chr> <dbl> <dbl> <dbl>
1 e 78.2 2 1
2 EE 33.2 44 1
3 S 15 24 1
4 Q 20.4 5.6 2
5 W 14.6 4 1
答案 1 :(得分:0)
不确定您是否要从特定的程序包中寻找解决方案,可以使用dplyr
程序包来实现,其中我们使用group_by
的{{1}}列和category
结果根据我们的需要。
这是示例代码。
代码:
summarise
输出:
f %>% group_by(category) %>%
summarise(sumDD = sum(DD), MeanCC = mean(CC), CountDistinctFF = length(unique(FF)))
答案 2 :(得分:0)
作为对安东尼斯的回应的补充方式,后来出现了一个错误。经过调查,我发现软件包之间存在冲突,特别是dplyr和SparkR。
这可以通过安装tidyverse软件包并按如下所示调用命令来解决:
>library(tidyverse)
>df0=df%>%dplyr::group_by(category)%>%dplyr::summarize(sumDD=sum(DD,na.rm=T),MeanCC=mean(CC,na.rm=T),CountDistinctFF=n_distinct(FF))
>glimpse(df0)
Observations: ??
Variables: 4
$ category <chr> "e", "EE", "S", "Q", "W"
$ sumDD <dbl> 78.20, 33.20, 15.00, 20.40, 14.55
$ MeanCC <dbl> 2.0, 44.0, 24.0, 5.6, 4.0
$ CountDistinctFF <dbl> 1, 1, 1, 2, 1