SparkChain

laktory.models.SparkChain

Bases: BaseChain

The SparkChain class defines a series of Spark transformation to be applied to a dataframe. Each transformation is expressed as a node (SparkChainNode object) that, upon execution, returns a new dataframe. Each node is executed sequentially in the provided order. A node may also be another SparkChain.

ATTRIBUTE	DESCRIPTION
dataframe_backend	Differentiator to select dataframe chain type TYPE: Literal['SPARK']
nodes	The list of transformations to be executed. TYPE: list[Union[SparkChainNode, SparkChain]]

Examples:

import pandas as pd
from laktory import models

df0 = spark.createDataFrame(pd.DataFrame({"x": [1, 2, 3]}))

# Build Chain
sc = models.SparkChain(
    nodes=[
        {
            "with_column": {
                "name": "cos_x",
                "type": "double",
                "expr": "F.cos('x')",
            },
        },
        {
            "nodes": [
                {
                    "func_name": "withColumnRenamed",
                    "func_args": [
                        "x",
                        "x_tmp",
                    ],
                },
                {
                    "with_column": {
                        "name": "x2",
                        "type": "double",
                        "expr": "F.sqrt('x_tmp')",
                    },
                },
            ],
        },
        {
            "func_name": "drop",
            "func_args": [
                "x_tmp",
            ],
        },
    ]
)

# Execute Chain
df = sc.execute(df0)

# Print result
print(df.toPandas().to_string())
'''
      cos_x        x2
0  0.540302  1.000000
1 -0.416147  1.414214
2 -0.989992  1.732051
'''

--

laktory.models.SparkChainNode

Bases: BaseChainNode

PolarsChain node that output a dataframe upon execution. As a convenience, with_column argument can be specified to create a new column from a spark or sql expression. Each node is executed sequentially in the provided order. A node may also be another Spark Chain.

ATTRIBUTE	DESCRIPTION
func_args	List of arguments to be passed to the spark function. If the function expects a spark column, its string representation can be provided with support for col, lit, expr and F.. TYPE: list[Union[Any]]
func_kwargs	List of keyword arguments to be passed to the spark function. If the function expects a spark column, its string representation can be provided with support for col, lit, expr and F.. TYPE: dict[str, Union[Any]]
func_name	Name of the spark function to build the dataframe. Mutually exclusive to sql_expr and with_column. TYPE: Union[str, None]
sql_expr	SQL Expression using {df} to reference upstream dataframe and defining how to build the output dataframe. Mutually exclusive to func_name and with_column. Other pipeline nodes can also be referenced using {nodes.node_name}. TYPE: Union[str, None]
with_column	Syntactic sugar for adding a column. Mutually exclusive to func_name and sql_expr. TYPE: Union[ChainNodeColumn, None]
with_columns	Syntactic sugar for adding columns. Mutually exclusive to func_name and sql_expr. TYPE: Union[list[ChainNodeColumn], None]

Examples:

import pandas as pd
from laktory import models

df0 = spark.createDataFrame(pd.DataFrame({"x": [1, 2, 2, 3]}))

node = models.SparkChainNode(
    with_column={
        "name": "cosx",
        "type": "double",
        "expr": "F.cos('x')",
    },
)
df = node.execute(df0)

node = models.SparkChainNode(
    with_column={
        "name": "xy",
        "type": "double",
        "expr": "F.coalesce('x')",
    },
)
df = node.execute(df)

print(df.toPandas().to_string())
'''
   x      cosx   xy
0  1  0.540302  1.0
1  2 -0.416147  2.0
2  2 -0.416147  2.0
3  3 -0.989992  3.0
'''

node = models.SparkChainNode(
    func_name="drop_duplicates",
    func_args=[["x"]],
)
df = node.execute(df)

print(df.toPandas().to_string())
'''
   x      cosx   xy
0  1  0.540302  1.0
1  2 -0.416147  2.0
2  3 -0.989992  3.0
'''

METHOD	DESCRIPTION
execute	Execute spark chain node

Functions

execute

execute(df, udfs=None)

Execute spark chain node

PARAMETER	DESCRIPTION
df	Input dataframe TYPE: SparkDataFrame
udfs	User-defined functions TYPE: list[Callable[[...], Union[SparkColumn, SparkDataFrame]]] DEFAULT: None

RETURNS	DESCRIPTION
Output dataframe

Source code in laktory/models/transformers/sparkchainnode.py

def execute(
    self,
    df: SparkDataFrame,
    udfs: list[Callable[[...], Union[SparkColumn, SparkDataFrame]]] = None,
) -> Union[SparkDataFrame]:
    """
    Execute spark chain node

    Parameters
    ----------
    df:
        Input dataframe
    udfs:
        User-defined functions

    Returns
    -------
        Output dataframe
    """
    from laktory.spark import DATATYPES_MAP

    if udfs is None:
        udfs = []
    udfs = {f.__name__: f for f in udfs}

    # Build Columns
    if self._with_columns:
        for column in self._with_columns:
            logger.info(
                f"Building column {column.name} as {column.expr or column.sql_expr}"
            )
            _col = column.eval(udfs=udfs, dataframe_backend="SPARK")
            if column.type:
                _col = _col.cast(DATATYPES_MAP[column.type])
            df = df.withColumns({column.name: _col})
        return df

    # From SQL expression
    if self.sql_expr:
        logger.info(f"DataFrame {self.id} as \n{self.sql_expr.strip()}")
        return self.parsed_sql_expr.eval(df)

    # Get Function
    func_name = self.func_name
    if self.func_name is None:
        raise ValueError(
            "`func_name` must be specified if `sql_expr` is not specified"
        )

    # Get from UDFs
    f = udfs.get(func_name, None)

    # Get from built-in spark and spark extension (including Laktory) functions
    input_df = True
    if f is None:
        # Get function from namespace extension
        if "." in func_name:
            input_df = False
            vals = func_name.split(".")
            f = getattr(getattr(df, vals[0]), vals[1], None)
        else:
            f = getattr(type(df), func_name, None)

    if f is None:
        raise ValueError(f"Function {func_name} is not available")

    _args = self.parsed_func_args
    _kwargs = self.parsed_func_kwargs

    # Build log
    func_log = f"{func_name}("
    func_log += ",".join([a.signature() for a in _args])
    func_log += ",".join([f"{k}={a.signature()}" for k, a in _kwargs.items()])
    func_log += ")"
    logger.info(f"DataFrame {self.id} as {func_log}")

    # Build args
    args = []
    for i, _arg in enumerate(_args):
        args += [_arg.eval(spark=df.sparkSession)]

    # Build kwargs
    kwargs = {}
    for k, _arg in _kwargs.items():
        kwargs[k] = _arg.eval(spark=df.sparkSession)

    # Call function
    if input_df:
        df = f(df, *args, **kwargs)
    else:
        df = f(*args, **kwargs)

    return df

--

laktory.models.SparkChainNodeSQLExpr

Bases: BaseChainNodeSQLExpr

Chain node SQL expression

ATTRIBUTE	DESCRIPTION
expr	SQL expression TYPE: str