Skip to content

Arrays

A collection of classes for performing various transformations on arrays in PySpark.

These transformations include operations such as removing duplicates, exploding arrays into separate rows, reversing the order of elements, sorting elements, removing certain values, and calculating aggregate statistics like minimum, maximum, sum, mean, and median.

Concept
  • Every transformation in this module is implemented as a class that inherits from the ArrayTransformation class.
  • The ArrayTransformation class is a subclass of ColumnsTransformationWithTarget
  • The ArrayTransformation class implements the func method, which is used to define the transformation logic.
  • The func method takes a column as input and returns a Column object.
  • The Column object is a PySpark column that can be used to perform transformations on a DataFrame column.
  • The ArrayTransformation limits the data type of the transformation to array by setting the ColumnConfig class to run_for_all_data_type = [SparkDatatype.ARRAY] and limit_data_type = [SparkDatatype.ARRAY].
See Also

koheesio.spark.transformations.arrays.ArraySortAsc module-attribute #

ArraySortAsc = ArraySort

koheesio.spark.transformations.arrays.ArrayDistinct #

Remove duplicates from array

Example
ArrayDistinct(column="array_column")

filter_empty class-attribute instance-attribute #

filter_empty: bool = Field(
    default=True,
    description="Remove null, nan, and empty values from array. Default is True.",
)

func #

func(column: Column) -> Column
Source code in src/koheesio/spark/transformations/arrays.py
def func(self, column: Column) -> Column:
    _fn = F.array_distinct(column)

    # noinspection PyUnresolvedReferences
    element_type = self.column_type_of_col(column, None, False).elementType
    is_numeric = spark_data_type_is_numeric(element_type)

    if self.filter_empty:
        # Remove null values from array
        if spark_minor_version >= 3.4:
            # Run array_compact if spark version is 3.4 or higher
            # https://spark.apache.org/docs/latest/api/python/reference/pyspark.sql/api/pyspark.sql.functions.array_compact.html
            # pylint: disable=E0611
            from pyspark.sql.functions import array_compact as _array_compact

            _fn = _array_compact(_fn)
            # pylint: enable=E0611
        else:
            # Otherwise, remove null from array using array_except
            _fn = F.array_except(_fn, F.array(F.lit(None)))

        # Remove nan or empty values from array (depends on the type of the elements in array)
        if is_numeric:
            # Remove nan from array (float/int/numbers)
            _fn = F.array_except(_fn, F.array(F.lit(float("nan")).cast(element_type)))
        else:
            # Remove empty values from array (string/text)
            _fn = F.array_except(_fn, F.array(F.lit(""), F.lit(" ")))

    return _fn

koheesio.spark.transformations.arrays.ArrayMax #

Return the maximum value in the array

Example
ArrayMax(column="array_column")

func #

func(column: Column) -> Column
Source code in src/koheesio/spark/transformations/arrays.py
def func(self, column: Column) -> Column:
    # Call for processing of nan values
    column = super().func(column)

    return F.array_max(column)

koheesio.spark.transformations.arrays.ArrayMean #

Return the mean of the values in the array.

Note: Only numeric values are supported for calculating the mean.

Example
ArrayMean(column="array_column", target_column="average")

func #

func(column: Column) -> Column

Calculate the mean of the values in the array

Source code in src/koheesio/spark/transformations/arrays.py
def func(self, column: Column) -> Column:
    """Calculate the mean of the values in the array"""
    # raise an error if the array contains non-numeric elements
    element_type = self.column_type_of_col(col=column, df=None, simple_return_mode=False).elementType

    if not spark_data_type_is_numeric(element_type):
        raise ValueError(
            f"{column = } contains non-numeric values. The array type is {element_type}. "
            f"Only numeric values are supported for calculating a mean."
        )

    _sum = ArraySum.from_step(self).func(column)
    # Call for processing of nan values
    column = super().func(column)
    _size = F.size(column)
    # return 0 if the size of the array is 0 to avoid division by zero
    return F.when(_size == 0, F.lit(0)).otherwise(_sum / _size)

koheesio.spark.transformations.arrays.ArrayMedian #

Return the median of the values in the array.

The median is the middle value in a sorted, ascending or descending, list of numbers.

  • If the size of the array is even, the median is the average of the two middle numbers.
  • If the size of the array is odd, the median is the middle number.

Note: Only numeric values are supported for calculating the median.

Example
ArrayMedian(column="array_column", target_column="median")

func #

func(column: Column) -> Column

Calculate the median of the values in the array

Source code in src/koheesio/spark/transformations/arrays.py
def func(self, column: Column) -> Column:
    """Calculate the median of the values in the array"""
    # Call for processing of nan values
    column = super().func(column)

    sorted_array = ArraySort.from_step(self).func(column)
    _size: Column = F.size(sorted_array)

    # Calculate the middle index. If the size is odd, PySpark discards the fractional part.
    # Use floor function to ensure the result is an integer
    middle: Column = F.floor((_size + 1) / 2).cast("int")

    # Define conditions
    is_size_zero: Column = _size == 0
    is_column_null: Column = column.isNull()
    is_size_even: Column = _size % 2 == 0

    # Define actions / responses
    # For even-sized arrays, calculate the average of the two middle elements
    average_of_middle_elements = (F.element_at(sorted_array, middle) + F.element_at(sorted_array, middle + 1)) / 2
    # For odd-sized arrays, select the middle element
    middle_element = F.element_at(sorted_array, middle)
    # In case the array is empty, return either None or 0
    none_value = F.lit(None)
    zero_value = F.lit(0)

    median = (
        # Check if the size of the array is 0
        F.when(
            is_size_zero,
            # If the size of the array is 0 and the column is null, return None
            # If the size of the array is 0 and the column is not null, return 0
            F.when(is_column_null, none_value).otherwise(zero_value),
        ).otherwise(
            # If the size of the array is not 0, calculate the median
            F.when(is_size_even, average_of_middle_elements).otherwise(middle_element)
        )
    )

    return median

koheesio.spark.transformations.arrays.ArrayMin #

Return the minimum value in the array

Example
ArrayMin(column="array_column")

func #

func(column: Column) -> Column
Source code in src/koheesio/spark/transformations/arrays.py
def func(self, column: Column) -> Column:
    return F.array_min(column)

koheesio.spark.transformations.arrays.ArrayNullNanProcess #

Process an array by removing NaN and/or NULL values from elements.

Parameters:

Name Type Description Default
keep_nan bool

Whether to keep NaN values in the array. If set to True, the NaN values will be kept in the array.

False
keep_null bool

Whether to keep NULL values in the array. If set to True, the NULL values will be kept in the array.

False

Returns:

Name Type Description
column Column

The processed column with NaN and/or NULL values removed from elements.

Examples:

>>> input_data = [(1, [1.1, 2.1, 4.1, float("nan")])]
>>> input_schema = StructType([StructField("id", IntegerType(), True),
    StructField("array_float", ArrayType(FloatType()), True),
])
>>> spark = SparkSession.builder.getOrCreate()
>>> df = spark.createDataFrame(input_data, schema=input_schema)
>>> transformer = ArrayNumericNanProcess(column="array_float", keep_nan=False)
>>> transformer.transform(df)
>>> print(transformer.output.df.collect()[0].asDict()["array_float"])
[1.1, 2.1, 4.1]

>>> input_data = [(1, [1.1, 2.2, 4.1, float("nan")])]
>>> input_schema = StructType([StructField("id", IntegerType(), True),
    StructField("array_float", ArrayType(FloatType()), True),
])
>>> spark = SparkSession.builder.getOrCreate()
>>> df = spark.createDataFrame(input_data, schema=input_schema)
>>> transformer = ArrayNumericNanProcess(column="array_float", keep_nan=True)
>>> transformer.transform(df)
>>> print(transformer.output.df.collect()[0].asDict()["array_float"])
[1.1, 2.1, 4.1, nan]

keep_nan class-attribute instance-attribute #

keep_nan: bool = Field(
    False,
    description="Whether to keep nan values in the array. Default is False. If set to True, the nan values will be kept in the array.",
)

keep_null class-attribute instance-attribute #

keep_null: bool = Field(
    False,
    description="Whether to keep null values in the array. Default is False. If set to True, the null values will be kept in the array.",
)

func #

func(column: Column) -> Column

Process the given column by removing NaN and/or NULL values from elements.

Parameters:

column : Column The column to be processed.

Returns:

column : Column The processed column with NaN and/or NULL values removed from elements.

Source code in src/koheesio/spark/transformations/arrays.py
def func(self, column: Column) -> Column:
    """
    Process the given column by removing NaN and/or NULL values from elements.

    Parameters:
    -----------
    column : Column
        The column to be processed.

    Returns:
    --------
    column : Column
        The processed column with NaN and/or NULL values removed from elements.
    """

    def apply_logic(x: Column):
        if self.keep_nan is False and self.keep_null is False:
            logic = x.isNotNull() & ~F.isnan(x)
        elif self.keep_nan is False:
            logic = ~F.isnan(x)
        elif self.keep_null is False:
            logic = x.isNotNull()

        return logic

    if self.keep_nan is False or self.keep_null is False:
        column = F.filter(column, apply_logic)

    return column

koheesio.spark.transformations.arrays.ArrayRemove #

Remove a certain value from the array

Parameters:

Name Type Description Default
keep_nan bool

Whether to keep NaN values in the array. If set to True, the NaN values will be kept in the array.

False
keep_null bool

Whether to keep NULL values in the array. If set to True, the NULL values will be kept in the array.

False
Example
ArrayRemove(column="array_column", value="value_to_remove")

make_distinct class-attribute instance-attribute #

make_distinct: bool = Field(
    default=False,
    description="Whether to remove duplicates from the array.",
)

value class-attribute instance-attribute #

value: Any = Field(
    default=None,
    description="The value to remove from the array.",
)

func #

func(column: Column) -> Column
Source code in src/koheesio/spark/transformations/arrays.py
def func(self, column: Column) -> Column:
    value = self.value

    column = super().func(column)

    def filter_logic(x: Column, _val: Any):
        if self.keep_null and self.keep_nan:
            logic = (x != F.lit(_val)) | x.isNull() | F.isnan(x)
        elif self.keep_null:
            logic = (x != F.lit(_val)) | x.isNull()
        elif self.keep_nan:
            logic = (x != F.lit(_val)) | F.isnan(x)
        else:
            logic = x != F.lit(_val)

        return logic

    # Check if the value is iterable (i.e., a list, tuple, or set)
    if isinstance(value, (list, tuple, set)):
        result = reduce(lambda res, val: F.filter(res, lambda x: filter_logic(x, val)), value, column)
    else:
        # If the value is not iterable, simply remove the value from the array
        result = F.filter(column, lambda x: filter_logic(x, value))

    if self.make_distinct:
        result = F.array_distinct(result)

    return result

koheesio.spark.transformations.arrays.ArrayReverse #

Reverse the order of elements in the array

Example
ArrayReverse(column="array_column")

func #

func(column: Column) -> Column
Source code in src/koheesio/spark/transformations/arrays.py
def func(self, column: Column) -> Column:
    return F.reverse(column)

koheesio.spark.transformations.arrays.ArraySort #

Sort the elements in the array

By default, the elements are sorted in ascending order. To sort the elements in descending order, set the reverse parameter to True.

Example
ArraySort(column="array_column")

reverse class-attribute instance-attribute #

reverse: bool = Field(
    default=False,
    description="Sort the elements in the array in a descending order. Default is False.",
)

func #

func(column: Column) -> Column
Source code in src/koheesio/spark/transformations/arrays.py
def func(self, column: Column) -> Column:
    column = F.array_sort(column)
    if self.reverse:
        # Reverse the order of elements in the array
        column = ArrayReverse.from_step(self).func(column)
    return column

koheesio.spark.transformations.arrays.ArraySortDesc #

Sort the elements in the array in descending order

reverse class-attribute instance-attribute #

reverse: bool = True

koheesio.spark.transformations.arrays.ArraySum #

Return the sum of the values in the array

Parameters:

Name Type Description Default
keep_nan bool

Whether to keep NaN values in the array. If set to True, the NaN values will be kept in the array.

False
keep_null bool

Whether to keep NULL values in the array. If set to True, the NULL values will be kept in the array.

False
Example
ArraySum(column="array_column")

func #

func(column: Column) -> Column

Using the aggregate function to sum the values in the array

Source code in src/koheesio/spark/transformations/arrays.py
def func(self, column: Column) -> Column:
    """Using the `aggregate` function to sum the values in the array"""
    # raise an error if the array contains non-numeric elements
    element_type = self.column_type_of_col(column, None, False).elementType
    if not spark_data_type_is_numeric(element_type):
        raise ValueError(
            f"{column = } contains non-numeric values. The array type is {element_type}. "
            f"Only numeric values are supported for summing."
        )

    # remove na values from array.
    column = super().func(column)

    # Using the `aggregate` function to sum the values in the array by providing the initial value as 0.0 and the
    # lambda function to add the elements together. Pyspark will automatically infer the type of the initial value
    # making 0.0 valid for both integer and float types.
    initial_value = F.lit(0.0)
    return F.aggregate(column, initial_value, lambda accumulator, x: accumulator + x)

koheesio.spark.transformations.arrays.ArrayTransformation #

Base class for array transformations

ColumnConfig #

Set the data type of the Transformation to array

limit_data_type class-attribute instance-attribute #

limit_data_type = [ARRAY]

run_for_all_data_type class-attribute instance-attribute #

run_for_all_data_type = [ARRAY]

func #

func(column: Column) -> Column
Source code in src/koheesio/spark/transformations/arrays.py
def func(self, column: Column) -> Column:
    raise NotImplementedError("This is an abstract class")

koheesio.spark.transformations.arrays.Explode #

Explode the array into separate rows

Example
Explode(column="array_column")

distinct class-attribute instance-attribute #

distinct: bool = Field(
    False,
    description="Remove duplicates from the exploded array. Default is False.",
)

preserve_nulls class-attribute instance-attribute #

preserve_nulls: bool = Field(
    True,
    description="Preserve rows with null values in the exploded array by using explode_outer instead of explode.Default is True.",
)

func #

func(column: Column) -> Column
Source code in src/koheesio/spark/transformations/arrays.py
def func(self, column: Column) -> Column:
    if self.distinct:
        column = ArrayDistinct.from_step(self).func(column)
    return F.explode_outer(column) if self.preserve_nulls else F.explode(column)

koheesio.spark.transformations.arrays.ExplodeDistinct #

Explode the array into separate rows while removing duplicates and empty values

Example
ExplodeDistinct(column="array_column")

distinct class-attribute instance-attribute #

distinct: bool = True