Changing nested column names using SparklyR in R

问题

I have referred to all the links mentioned here:

1) Link-1 2) Link-2 3) Link-3 4) Link-4

Following R code has been written by using Sparklyr Package. It reads huge JSON file and creates database schema.

sc <- spark_connect(master = "local", config = conf, version = '2.2.0') # Connection
sample_tbl <- spark_read_json(sc,name="example",path="example.json", header = TRUE, 
                              memory = FALSE, overwrite = TRUE) # reads JSON file
sample_tbl <- sdf_schema_viewer(sample_tbl) # to create db schema
df <- tbl(sc,"example") # to create lookup table

It has created following database schema

Now,

If I rename first level column, then it works.

For example,

df %>% rename(ent = entities)

But when I run 2nd deep level nested column then it doesn't rename.

df %>% rename(e_hashtags = entities.hashtags)

It shows error:

Error in .f(.x[[i]], ...) : object 'entities.hashtags' not found

Question

My question is, how to rename 3rd to 4th deep level nested column also?

Please refer database schema mentioned above.

回答1:

Spark as such doesn't support renaming individual nested fields. You have to either cast or rebuild a whole structure. For simplicity let's assume that data looks as follows:

cat('{"contributors": "foo", "coordinates": "bar", "entities": {"hashtags": ["foo", "bar"], "media": "missing"}}',  file = "/tmp/example.json")
df <- spark_read_json(sc, "df", "/tmp/example.json", overwrite=TRUE)

df %>% spark_dataframe() %>% invoke("schema") %>% invoke("treeString") %>% cat()

root
 |-- contributors: string (nullable = true)
 |-- coordinates: string (nullable = true)
 |-- entities: struct (nullable = true)
 |    |-- hashtags: array (nullable = true)
 |    |    |-- element: string (containsNull = true)
 |    |-- media: string (nullable = true)

with simple string representation:

df %>% 
  spark_dataframe() %>% 
  invoke("schema") %>% 
  invoke("simpleString") %>% 
  cat(sep = "\n")

struct<contributors:string,coordinates:string,entities:struct<hashtags:array<string>,media:string>>

With cast you have to define expression using matching type description:

expr_cast <- invoke_static(
  sc, "org.apache.spark.sql.functions", "expr",
  "CAST(entities AS struct<e_hashtags:array<string>,media:string>)"
)

df_cast <- df %>% 
  spark_dataframe() %>% 
  invoke("withColumn", "entities", expr_cast) %>% 
  sdf_register()

df_cast %>% spark_dataframe() %>% invoke("schema") %>% invoke("treeString") %>% cat()

root
 |-- contributors: string (nullable = true)
 |-- coordinates: string (nullable = true)
 |-- entities: struct (nullable = true)
 |    |-- e_hashtags: array (nullable = true)
 |    |    |-- element: string (containsNull = true)
 |    |-- media: string (nullable = true)

To rebuild structure you have to match all components:

expr_struct <- invoke_static(
  sc, "org.apache.spark.sql.functions", "expr",
  "struct(entities.hashtags AS e_hashtags, entities.media)"
)

df_struct <- df %>% 
  spark_dataframe() %>% 
  invoke("withColumn", "entities", expr_struct) %>% 
  sdf_register()

df_struct %>% spark_dataframe() %>% invoke("schema") %>% invoke("treeString") %>% cat()

root
 |-- contributors: string (nullable = true)
 |-- coordinates: string (nullable = true)
 |-- entities: struct (nullable = false)
 |    |-- e_hashtags: array (nullable = true)
 |    |    |-- element: string (containsNull = true)
 |    |-- media: string (nullable = true)

来源：https://stackoverflow.com/questions/52263836/changing-nested-column-names-using-sparklyr-in-r

标签

apache-spark

sparklyr