How to flatten a struct in a Spark dataframe?

Question

I have a dataframe with the following structure:

 |-- data: struct (nullable = true)
 |    |-- id: long (nullable = true)
 |    |-- keyNote: struct (nullable = true)
 |    |    |-- key: string (nullable = true)
 |    |    |-- note: string (nullable = true)
 |    |-- details: map (nullable = true)
 |    |    |-- key: string
 |    |    |-- value: string (valueContainsNull = true)

How it is possible to flatten the structure and create a new dataframe:

     |-- id: long (nullable = true)
     |-- keyNote: struct (nullable = true)
     |    |-- key: string (nullable = true)
     |    |-- note: string (nullable = true)
     |-- details: map (nullable = true)
     |    |-- key: string
     |    |-- value: string (valueContainsNull = true)

Is there something like explode, but for structs?

Answer 1

This should work in Spark 1.6 or later:

df.select(df.col("data.*"))

or

df.select(df.col("data.id"), df.col("data.keyNote"), df.col("data.details"))

Answer 2

Here is function that is doing what you want and that can deal with multiple nested columns containing columns with same name:

import pyspark.sql.functions as F

def flatten_df(nested_df):
    flat_cols = [c[0] for c in nested_df.dtypes if c[1][:6] != 'struct']
    nested_cols = [c[0] for c in nested_df.dtypes if c[1][:6] == 'struct']

    flat_df = nested_df.select(flat_cols +
                               [F.col(nc+'.'+c).alias(nc+'_'+c)
                                for nc in nested_cols
                                for c in nested_df.select(nc+'.*').columns])
    return flat_df

Before:

root
 |-- x: string (nullable = true)
 |-- y: string (nullable = true)
 |-- foo: struct (nullable = true)
 |    |-- a: float (nullable = true)
 |    |-- b: float (nullable = true)
 |    |-- c: integer (nullable = true)
 |-- bar: struct (nullable = true)
 |    |-- a: float (nullable = true)
 |    |-- b: float (nullable = true)
 |    |-- c: integer (nullable = true)

After:

root
 |-- x: string (nullable = true)
 |-- y: string (nullable = true)
 |-- foo_a: float (nullable = true)
 |-- foo_b: float (nullable = true)
 |-- foo_c: integer (nullable = true)
 |-- bar_a: float (nullable = true)
 |-- bar_b: float (nullable = true)
 |-- bar_c: integer (nullable = true)

Answer 3

This flatten_df version flattens the dataframe at every layer level, using a stack to avoid recursive calls:

from pyspark.sql.functions import col


def flatten_df(nested_df):
    stack = [((), nested_df)]
    columns = []

    while len(stack) > 0:
        parents, df = stack.pop()

        flat_cols = [
            col(".".join(parents + (c[0],))).alias("_".join(parents + (c[0],)))
            for c in df.dtypes
            if c[1][:6] != "struct"
        ]

        nested_cols = [
            c[0]
            for c in df.dtypes
            if c[1][:6] == "struct"
        ]

        columns.extend(flat_cols)

        for nested_col in nested_cols:
            projected_df = df.select(nested_col + ".*")
            stack.append((parents + (nested_col,), projected_df))

    return nested_df.select(columns)

Example:

from pyspark.sql.types import StringType, StructField, StructType


schema = StructType([
    StructField("some", StringType()),

    StructField("nested", StructType([
        StructField("nestedchild1", StringType()),
        StructField("nestedchild2", StringType())
    ])),

    StructField("renested", StructType([
        StructField("nested", StructType([
            StructField("nestedchild1", StringType()),
            StructField("nestedchild2", StringType())
        ]))
    ]))
])

data = [
    {
        "some": "value1",
        "nested": {
            "nestedchild1": "value2",
            "nestedchild2": "value3",
        },
        "renested": {
            "nested": {
                "nestedchild1": "value4",
                "nestedchild2": "value5",
            }
        }
    }
]

df = spark.createDataFrame(data, schema)
flat_df = flatten_df(df)
print(flat_df.collect())

Prints:

[Row(some=u'value1', renested_nested_nestedchild1=u'value4', renested_nested_nestedchild2=u'value5', nested_nestedchild1=u'value2', nested_nestedchild2=u'value3')]

Answer 4

For Spark 2.4.5,

while,df.select(df.col("data.*")) will give you org.apache.spark.sql.AnalysisException: No such struct field * in exception

this will work:-

df.select($"data.*")

Answer 5

I generalized the solution from stecos a bit more so the flattening can be done on more than two struct layers deep:

def flatten_df(nested_df, layers):
    flat_cols = []
    nested_cols = []
    flat_df = []

    flat_cols.append([c[0] for c in nested_df.dtypes if c[1][:6] != 'struct'])
    nested_cols.append([c[0] for c in nested_df.dtypes if c[1][:6] == 'struct'])

    flat_df.append(nested_df.select(flat_cols[0] +
                               [col(nc+'.'+c).alias(nc+'_'+c)
                                for nc in nested_cols[0]
                                for c in nested_df.select(nc+'.*').columns])
                  )
    for i in range(1, layers):
        print (flat_cols[i-1])
        flat_cols.append([c[0] for c in flat_df[i-1].dtypes if c[1][:6] != 'struct'])
        nested_cols.append([c[0] for c in flat_df[i-1].dtypes if c[1][:6] == 'struct'])

        flat_df.append(flat_df[i-1].select(flat_cols[i] +
                                [col(nc+'.'+c).alias(nc+'_'+c)
                                    for nc in nested_cols[i]
                                    for c in flat_df[i-1].select(nc+'.*').columns])
        )

    return flat_df[-1]

just call with:

my_flattened_df = flatten_df(my_df_having_nested_structs, 3)

(second parameter is the level of layers to be flattened, in my case it's 3)

Answer 6

You can use this approach if, you have to covert only struct types. I would not suggest converting the array, as it could lead to duplicate records.

from pyspark.sql.functions import col
from pyspark.sql.types import StructType


def flatten_schema(schema, prefix=""):
    return_schema = []
    for field in schema.fields:
        if isinstance(field.dataType, StructType):
            if prefix:
                return_schema = return_schema + flatten_schema(field.dataType, "{}.{}".format(prefix, field.name))
            else:
                return_schema = return_schema + flatten_schema(field.dataType, field.name)
        else:
            if prefix:
                field_path = "{}.{}".format(prefix, field.name)
                return_schema.append(col(field_path).alias(field_path.replace(".", "_")))
            else:
                return_schema.append(field.name)
    return return_schema

You can use it as

new_schema = flatten_schema(df.schema)
df1 = df.select(se)
df1.show()

Answer 7

PySpark solution to flatten nested df with both struct and array types with any level of depth. This is improved on this: https://stackoverflow.com/a/56533459/7131019

from pyspark.sql.types import *
from pyspark.sql import functions as f

def flatten_structs(nested_df):
    stack = [((), nested_df)]
    columns = []

    while len(stack) > 0:
        
        parents, df = stack.pop()
        
        array_cols = [
            c[0]
            for c in df.dtypes
            if c[1][:5] == "array"
        ]
        
        flat_cols = [
            f.col(".".join(parents + (c[0],))).alias("_".join(parents + (c[0],)))
            for c in df.dtypes
            if c[1][:6] != "struct"
        ]

        nested_cols = [
            c[0]
            for c in df.dtypes
            if c[1][:6] == "struct"
        ]
        
        columns.extend(flat_cols)

        for nested_col in nested_cols:
            projected_df = df.select(nested_col + ".*")
            stack.append((parents + (nested_col,), projected_df))
        
    return nested_df.select(columns)

def flatten_array_struct_df(df):
    
    array_cols = [
            c[0]
            for c in df.dtypes
            if c[1][:5] == "array"
        ]
    
    while len(array_cols) > 0:
        
        for array_col in array_cols:
            
            cols_to_select = [x for x in df.columns if x != array_col ]
            
            df = df.withColumn(array_col, f.explode(f.col(array_col)))
            
        df = flatten_structs(df)
        
        array_cols = [
            c[0]
            for c in df.dtypes
            if c[1][:5] == "array"
        ]
    return df

flat_df = flatten_array_struct_df(df)

Answer 8

An easy way is to use SQL, you could build a SQL query string to alias nested column as flat ones.

• Retrieve data-frame schema (df.schema())
• Transform schema to SQL (for (field : schema().fields()) ...

• Query:

  val newDF = sqlContext.sql("SELECT " + sqlGenerated + " FROM source")
  

Here is an example in Java.

(I prefer SQL way, so you can easily test it on Spark-shell and it's cross-language).

Answer 9

below worked for me in spark sql

import org.apache.spark.sql._
import org.apache.hadoop.conf.Configuration
import org.apache.hadoop.fs._
import org.apache.http.client.methods.HttpGet
import org.apache.http.impl.client.DefaultHttpClient
import org.apache.spark.sql.{DataFrame, SaveMode, SparkSession}
import org.apache.spark.sql.functions._
import org.apache.spark.sql.types.StructType
import org.apache.log4j.{Level, Logger}
import org.apache.spark.rdd.RDD
import org.apache.spark.sql.types._
import org.apache.spark.storage.StorageLevel
import org.apache.spark.sql.functions.{explode, expr, posexplode, when}

object StackOverFlowQuestion {
  def main(args: Array[String]): Unit = {

    val logger = Logger.getLogger("FlattenTest")
    Logger.getLogger("org").setLevel(Level.WARN)
    Logger.getLogger("akka").setLevel(Level.WARN)

    val spark = SparkSession.builder()
      .appName("FlattenTest")
      .config("spark.sql.warehouse.dir", "C:\\Temp\\hive")
      .master("local[2]")
      //.enableHiveSupport()
      .getOrCreate()
    import spark.implicits._

    val stringTest =
      """{
                               "total_count": 123,
                               "page_size": 20,
                               "another_id": "gdbfdbfdbd",
                               "sen": [{
                                "id": 123,
                                "ses_id": 12424343,
                                "columns": {
                                    "blah": "blah",
                                    "count": 1234
                                },
                                "class": {},
                                "class_timestamps": {},
                                "sentence": "spark is good"
                               }]
                            }
                             """
    val result = List(stringTest)
    val githubRdd=spark.sparkContext.makeRDD(result)
    val gitHubDF=spark.read.json(githubRdd)
    gitHubDF.show()
    gitHubDF.printSchema()

    gitHubDF.registerTempTable("JsonTable")

   spark.sql("with cte as" +
      "(" +
      "select explode(sen) as senArray  from JsonTable" +
      "), cte_2 as" +
      "(" +
      "select senArray.ses_id,senArray.ses_id,senArray.columns.* from cte" +
      ")" +
      "select * from cte_2"
    ).show()

    spark.stop()
}

}

output:-

+----------+---------+--------------------+-----------+
|another_id|page_size|                 sen|total_count|
+----------+---------+--------------------+-----------+
|gdbfdbfdbd|       20|[[[blah, 1234], 1...|        123|
+----------+---------+--------------------+-----------+

root
 |-- another_id: string (nullable = true)
 |-- page_size: long (nullable = true)
 |-- sen: array (nullable = true)
 |    |-- element: struct (containsNull = true)
 |    |    |-- columns: struct (nullable = true)
 |    |    |    |-- blah: string (nullable = true)
 |    |    |    |-- count: long (nullable = true)
 |    |    |-- id: long (nullable = true)
 |    |    |-- sentence: string (nullable = true)
 |    |    |-- ses_id: long (nullable = true)
 |-- total_count: long (nullable = true)

+--------+--------+----+-----+
|  ses_id|  ses_id|blah|count|
+--------+--------+----+-----+
|12424343|12424343|blah| 1234|
+--------+--------+----+-----+

Answer 10

A little more compact and efficient implementation:

No need to create list and iterate on them. You "act" on fields based on their type (if structures or not).

def flatten_df(nested_df):
    stack = [((), nested_df)]
    columns = []
    while len(stack) > 0:
        parents, df = stack.pop()
        for column_name, column_type in df.dtypes:
            if column_type[:6] == "struct":
                projected_df = df.select(column_name + ".*")
                stack.append((parents + (column_name,), projected_df))
            else:
                columns.append(col(".".join(parents + (column_name,))).alias("_".join(parents + (column_name,))))
    return nested_df.select(columns)

Answer 11

This is for scala spark.

val totalMainArrayBuffer=collection.mutable.ArrayBuffer[String]()
def flatten_df_Struct(dfTemp:org.apache.spark.sql.DataFrame,dfTotalOuter:org.apache.spark.sql.DataFrame):org.apache.spark.sql.DataFrame=
{
//dfTemp.printSchema
val totalStructCols=dfTemp.dtypes.map(x => x.toString.substring(1,x.toString.size-1)).filter(_.split(",",2)(1).contains("Struct")) // in case i the column names come with the word Struct embedded in it
val mainArrayBuffer=collection.mutable.ArrayBuffer[String]()
for(totalStructCol <- totalStructCols)
{
val tempArrayBuffer=collection.mutable.ArrayBuffer[String]()
tempArrayBuffer+=s"${totalStructCol.split(",")(0)}.*"
//tempArrayBuffer.toSeq.toDF.show(false)
val columnsInside=dfTemp.selectExpr(tempArrayBuffer:_*).columns
for(column <- columnsInside)
mainArrayBuffer+=s"${totalStructCol.split(",")(0)}.${column} as ${totalStructCol.split(",")(0)}_${column}"
//mainArrayBuffer.toSeq.toDF.show(false)
}
//dfTemp.selectExpr(mainArrayBuffer:_*).printSchema
val nonStructCols=dfTemp.selectExpr(mainArrayBuffer:_*).dtypes.map(x => x.toString.substring(1,x.toString.size-1)).filter(!_.split(",",2)(1).contains("Struct")) // in case i the column names come with the word Struct embedded in it
for (nonStructCol <- nonStructCols)
totalMainArrayBuffer+=s"${nonStructCol.split(",")(0).replace("_",".")} as ${nonStructCol.split(",")(0)}" // replacing _ by . in origial select clause if it's an already nested column 
dfTemp.selectExpr(mainArrayBuffer:_*).dtypes.map(x => x.toString.substring(1,x.toString.size-1)).filter(_.split(",",2)(1).contains("Struct")).size 
match {
case value if value ==0 => dfTotalOuter.selectExpr(totalMainArrayBuffer:_*)
case _ => flatten_df_Struct(dfTemp.selectExpr(mainArrayBuffer:_*),dfTotalOuter)
}
}


def flatten_df(dfTemp:org.apache.spark.sql.DataFrame):org.apache.spark.sql.DataFrame=
{
var totalArrayBuffer=collection.mutable.ArrayBuffer[String]()
val totalNonStructCols=dfTemp.dtypes.map(x => x.toString.substring(1,x.toString.size-1)).filter(!_.split(",",2)(1).contains("Struct")) // in case i the column names come with the word Struct embedded in it
for (totalNonStructCol <- totalNonStructCols)
totalArrayBuffer+=s"${totalNonStructCol.split(",")(0)}"
totalMainArrayBuffer.clear
flatten_df_Struct(dfTemp,dfTemp) // flattened schema is now in totalMainArrayBuffer 
totalArrayBuffer=totalArrayBuffer++totalMainArrayBuffer
dfTemp.selectExpr(totalArrayBuffer:_*)
}


flatten_df(dfTotal.withColumn("tempStruct",lit(5))).printSchema

File

{"num1":1,"num2":2,"bool1":true,"bool2":false,"double1":4.5,"double2":5.6,"str1":"a","str2":"b","arr1":[3,4,5],"map1":{"cool":1,"okay":2,"normal":3},"carInfo":{"Engine":{"Make":"sa","Power":{"IC":"900","battery":"165"},"Redline":"11500"} ,"Tyres":{"Make":"Pirelli","Compound":"c1","Life":"120"}}}
{"num1":3,"num2":4,"bool1":false,"bool2":false,"double1":4.2,"double2":5.5,"str1":"u","str2":"n","arr1":[6,7,9],"map1":{"fast":1,"medium":2,"agressive":3},"carInfo":{"Engine":{"Make":"na","Power":{"IC":"800","battery":"150"},"Redline":"10000"} ,"Tyres":{"Make":"Pirelli","Compound":"c2","Life":"100"}}}
{"num1":8,"num2":4,"bool1":true,"bool2":true,"double1":5.7,"double2":7.5,"str1":"t","str2":"k","arr1":[11,12,23],"map1":{"preserve":1,"medium":2,"fast":3},"carInfo":{"Engine":{"Make":"ta","Power":{"IC":"950","battery":"170"},"Redline":"12500"} ,"Tyres":{"Make":"Pirelli","Compound":"c3","Life":"80"}}}
{"num1":7,"num2":9,"bool1":false,"bool2":true,"double1":33.2,"double2":7.5,"str1":"b","str2":"u","arr1":[12,14,5],"map1":{"normal":1,"preserve":2,"agressive":3},"carInfo":{"Engine":{"Make":"pa","Power":{"IC":"920","battery":"160"},"Redline":"11800"} ,"Tyres":{"Make":"Pirelli","Compound":"c4","Life":"70"}}}

Before:

root
 |-- arr1: array (nullable = true)
 |    |-- element: long (containsNull = true)
 |-- bool1: boolean (nullable = true)
 |-- bool2: boolean (nullable = true)
 |-- carInfo: struct (nullable = true)
 |    |-- Engine: struct (nullable = true)
 |    |    |-- Make: string (nullable = true)
 |    |    |-- Power: struct (nullable = true)
 |    |    |    |-- IC: string (nullable = true)
 |    |    |    |-- battery: string (nullable = true)
 |    |    |-- Redline: string (nullable = true)
 |    |-- Tyres: struct (nullable = true)
 |    |    |-- Compound: string (nullable = true)
 |    |    |-- Life: string (nullable = true)
 |    |    |-- Make: string (nullable = true)
 |-- double1: double (nullable = true)
 |-- double2: double (nullable = true)
 |-- map1: struct (nullable = true)
 |    |-- agressive: long (nullable = true)
 |    |-- cool: long (nullable = true)
 |    |-- fast: long (nullable = true)
 |    |-- medium: long (nullable = true)
 |    |-- normal: long (nullable = true)
 |    |-- okay: long (nullable = true)
 |    |-- preserve: long (nullable = true)
 |-- num1: long (nullable = true)
 |-- num2: long (nullable = true)
 |-- str1: string (nullable = true)
 |-- str2: string (nullable = true

After:

root
 |-- arr1: array (nullable = true)
 |    |-- element: long (containsNull = true)
 |-- bool1: boolean (nullable = true)
 |-- bool2: boolean (nullable = true)
 |-- double1: double (nullable = true)
 |-- double2: double (nullable = true)
 |-- num1: long (nullable = true)
 |-- num2: long (nullable = true)
 |-- str1: string (nullable = true)
 |-- str2: string (nullable = true)
 |-- map1_agressive: long (nullable = true)
 |-- map1_cool: long (nullable = true)
 |-- map1_fast: long (nullable = true)
 |-- map1_medium: long (nullable = true)
 |-- map1_normal: long (nullable = true)
 |-- map1_okay: long (nullable = true)
 |-- map1_preserve: long (nullable = true)
 |-- carInfo_Engine_Make: string (nullable = true)
 |-- carInfo_Engine_Redline: string (nullable = true)
 |-- carInfo_Tyres_Compound: string (nullable = true)
 |-- carInfo_Tyres_Life: string (nullable = true)
 |-- carInfo_Tyres_Make: string (nullable = true)
 |-- carInfo_Engine_Power_IC: string (nullable = true)
 |-- carInfo_Engine_Power_battery: string (nullable = true)

Tried for 2 Levels, it worked

Answer 12

We used https://github.com/lvhuyen/SparkAid It works to any Level

from sparkaid import flatten

flatten(df_nested_B).printSchema()