Help Center/ DataArts Fabric/ Developer Guide/ Multimodal AI Data Lake/ Multimodal Data Processing/ User-Defined Functions/ Class UDF
Updated on 2025-12-19 GMT+08:00
View PDF
Share

Class UDF

A class UDF uses object-oriented programming to enable stateful processing logic. To implement it, you must define methods such as __init__ and __call__.

Class UDF Registration

When registering a class UDF, you pass a Python class where __init__, __call__, and __del__ have specific meanings (refer to the table below for details). Other class attributes and instance methods can be freely added by you without restrictions.

Table 1 Specific instance methods when registering a Python class

Python Class Method

Mandatory

Parameter

Description

Use Case

__init__(self, *args, **kwargs)

No

Parameters are passed through the with_arguments method. Only scalar values are allowed.

Constructor of the UDF.

Initializes UDF attributes (such as saving parameters, opening files, and establishing connections).

__call__(self, *args, **kwargs)

Yes

Parameters are passed through the UDF operator. Both scalar values and column names are allowed.

Entry point for the service logic of the UDF.

Processes data and performs calculations in the UDF.

__del__(self)

No

Does not support passing parameters.

Destructor of the UDF.

Clears UDF resources (for example, closing files and disconnecting networks).

Example

import ibis
import fabric_data as fabric
from fabric_data.udf import RegisterType

class Hello:
    def __init__(self, firstname: str, lastname: str): 
        self.firstname = firstname
        self.lastname = lastname

    def __call__(self, location: str) -> str:
        return f"Hello{self.firstname} {self.lastname}! Welcome to {location}"

con = ibis.fabric.connect(...)

# Explicitly register a UDF directly.
udf = con.udf.python.register(Hello, database="your-database", register_type=RegisterType.STAGED)
# Alternatively, register a UDF from a file.
udf = con.udf.python.register_from_file("your-current-file-path", "Hello", database="your-database", register_type=RegisterType.STAGED)

# Use the class UDF.
t = con.table(name="your-table", schema="your-schema")
expression = t.select(udf(t.city).with_arguments(firstname="Ethan", lastname="Carter"))

print(expression.execute())
Parent topic: User-Defined Functions