What is the equivalent of R's list() function in sparklyr?

问题

Below is a sample R code. I would like to do the same in sparklyr.

custTrans1 <- Pdt_table %>% 
  group_by(Main_CustomerID) %>% 
  summarise(Invoice = as.vector(list(Invoice_ID)),Industry = as.vector(list(Industry)))

where Pdt_table is spark data frame and Main_CustomerID, Invoice_ID and Industry are variables.

I would like to create list of the above variables and convert it to vector. How can I do it in sparklyr?

回答1:

You can use collect_list or collect_set:

set.seed(1)
df <- copy_to(
  sc, tibble(group = rep(c("a", "b"), 3), value = runif(6)),
  name = "df"
)

result <- df %>% group_by(group) %>% summarise(values = collect_list(value))
result

# Source:   lazy query [?? x 2]
# Database: spark_connection
  group values    
  <chr> <list>    
1 b     <list [3]>
2 a     <list [3]>

which is translated to the following query:

result %>% show_query()

<SQL>
SELECT `group`, COLLECT_LIST(`value`) AS `values`
FROM `df`
GROUP BY `group`

with corresponding execution plan:

result %>% optimizedPlan()

<jobj[213]>
  org.apache.spark.sql.catalyst.plans.logical.Aggregate
  Aggregate [group#259], [group#259, collect_list(value#260, 0, 0) AS values#345]
+- InMemoryRelation [group#259, value#260], true, 10000, StorageLevel(disk, memory, deserialized, 1 replicas), `df`
      +- Scan ExistingRDD[group#259,value#260]

and schema (with array<...> column):

root
 |-- group: string (nullable = true)
 |-- values: array (nullable = true)
 |    |-- element: double (containsNull = true)

Please keep in mind that:

Operation like this one is very expensive in a distributed system.
Depending on the data distribution might not be feasible.
Complex types are somewhat hard to handle in Spark in general, and sparklyr with it's tidy data focus, doesn't make things easier. To process the result efficiently you may require a Scala extension.

来源：https://stackoverflow.com/questions/50673835/what-is-the-equivalent-of-rs-list-function-in-sparklyr

标签

apache-spark

sparklyr