pyarrow.compute.register_vector_function#

pyarrow.compute.register_vector_function(func, function_name, function_doc, in_types, out_type, func_registry=None)#

Register a user-defined vector function.

This API is EXPERIMENTAL.

A vector function is a function that executes vector operations on arrays. Vector function is often used when compute doesnt fit other more specific types of functions (e.g., scalar and aggregate).

Parameters:

funccallable()

A callable implementing the user-defined function. The first argument is the context argument of type UdfContext. Then, it must take arguments equal to the number of in_types defined. It must return an Array or Scalar matching the out_type. It must return a Scalar if all arguments are scalar, else it must return an Array.

To define a varargs function, pass a callable that takes *args. The last in_type will be the type of all varargs arguments.

function_namestr

Name of the function. There should only be one function registered with this name in the function registry.

function_docdict

A dictionary object with keys summary (str), and description (str).

in_typesDict[str, DataType]

A dictionary mapping function argument names to their respective DataType. The argument names will be used to generate documentation for the function. The number of arguments specified here determines the function arity.

out_typeDataType

Output type of the function.

func_registryFunctionRegistry

Optional function registry to use instead of the default global one.

Examples

>>> import pyarrow as pa
>>> import pyarrow.compute as pc
>>>
>>> func_doc = {}
>>> func_doc["summary"] = "percent rank"
>>> func_doc["description"] = "compute percent rank"
>>>
>>> def list_flatten_udf(ctx, x):
...     return pc.list_flatten(x)
>>>
>>> func_name = "list_flatten_udf"
>>> in_types = {"array": pa.list_(pa.int64())}
>>> out_type = pa.int64()
>>> pc.register_vector_function(list_flatten_udf, func_name, func_doc,
...                   in_types, out_type)
>>>
>>> answer = pc.call_function(func_name, [pa.array([[1, 2], [3, 4]])])
>>> answer
<pyarrow.lib.Int64Array object at ...>
[
  1,
  2,
  3,
  4
]