Transform complex data types

While working with nested data types, Azure Databricks optimizes certain transformations out-of-the-box. The following code examples demonstrate patterns for working with complex and nested data types in Azure Databricks.

You can use dot notation (.) to access a nested field.

df.select("column_name.nested_field")

SELECT column_name.nested_field FROM table_name

Use the star operator (*) to select all fields within a given field.

df.select("column_name.*")

SELECT column_name.* FROM table_name

Use the struct() function to create a new nested field.

from pyspark.sql.functions import struct, col

df.select(struct(col("field_to_nest").alias("nested_field")).alias("column_name"))

SELECT struct(field_to_nest AS nested_field) AS column_name FROM table_name

Use the star operator (*) to nest all fields from a data source as a single column.

from pyspark.sql.functions import struct

df.select(struct("*").alias("column_name"))

SELECT struct(*) AS column_name FROM table_name

Use square brackets [] to select nested fields from a column.

from pyspark.sql.functions import col

df.select(col("column_name")["field_name"])

SELECT column_name["field_name"] FROM table_name

Use the explode() function to unpack values from ARRAY and MAP type columns.

ARRAY columns store values as a list. When unpacked with explode(), each value becomes a row in the output.

from pyspark.sql.functions import explode

df.select(explode("array_name").alias("column_name"))

SELECT explode(array_name) AS column_name FROM table_name

MAP columns store values as ordered key-value pairs. When unpacked with explode(), each key becomes a column and values become rows.

from pyspark.sql.functions import explode

df.select(explode("map_name").alias("column1_name", "column2_name"))

SELECT explode(map_name) AS (column1_name, column2_name) FROM table_name

Use the functions collect_list() or collect_set() to transform the values of a column into an array. collect_list() collects all values in the column, while collect_set() collects only unique values.

from pyspark.sql.functions import collect_list, collect_set

df.select(collect_list("column_name").alias("array_name"))
df.select(collect_set("column_name").alias("set_name"))

SELECT collect_list(column_name) AS array_name FROM table_name;
SELECT collect_set(column_name) AS set_name FROM table_name;

You can also use dot notation (.) to access fields in maps that are contained within an array. This returns an array of all values for the specified field.

Consider the following data structure:

{
  "column_name": [
    { "field1": 1, "field2": "a" },
    { "field1": 2, "field2": "b" }
  ]
}

You can return the values from field1 as an array with the following query:

df.select("column_name.field1")

SELECT column_name.field1 FROM table_name

Use the to_json function to convert a complex data type to JSON.

from pyspark.sql.functions import to_json

df.select(to_json("column_name").alias("json_name"))

SELECT to_json(column_name) AS json_name FROM table_name

To encode all contents of a query or DataFrame, combine this with struct(*).

from pyspark.sql.functions import to_json, struct

df.select(to_json(struct("*")).alias("json_name"))

SELECT to_json(struct(*)) AS json_name FROM table_name

Use the from_json function to convert JSON data to native complex data types.

from pyspark.sql.functions import from_json

schema = "column1 STRING, column2 DOUBLE"

df.select(from_json("json_name", schema).alias("column_name"))

SELECT from_json(json_name, "column1 STRING, column2 DOUBLE") AS column_name FROM table_name

The following notebooks provide examples for working with complex data types for Python, Scala, and SQL.

Get notebook

Get notebook

Get notebook