Implementation:Snorkel team Snorkel TransformationFunction Init

Knowledge Sources	Snorkel
Domains	Data_Augmentation, Data_Programming
Last Updated	2026-02-14 20:00 GMT

Overview

Concrete tool for defining data transformation functions that generate augmented examples, provided by the Snorkel library.

Description

The TransformationFunction class (extending Mapper) and the transformation_function decorator provide the interface for defining augmentation operations. A TF wraps a function that takes a data point and returns a modified copy or None.

The class inherits from Mapper which provides:

• Automatic deep copying of data points
• Field name mapping
• Preprocessor chaining
• Optional memoization

Usage

Import this class when defining augmentation operations. Use the decorator for simple TFs and the class form for complex transformations requiring field mappings.

Code Reference

Source Location

• Repository: snorkel
• File: snorkel/augmentation/tf.py
• Lines: L16-50

Signature

class TransformationFunction(Mapper):
    """Base class for transformation functions."""
    pass

class LambdaTransformationFunction(LambdaMapper):
    """TF defined from a simple function."""
    pass

class transformation_function:
    """Decorator to define a TransformationFunction."""
    # Returns LambdaTransformationFunction via lambda_mapper

Import

from snorkel.augmentation import transformation_function
# or
from snorkel.augmentation.tf import TransformationFunction

I/O Contract

Inputs

Name	Type	Required	Description
f	Callable	Yes	Function taking a data point, returning modified copy or None
name	Optional[str]	No	TF identifier (defaults to function name)
pre	Optional[List[BaseMapper]]	No	Preprocessing chain

Outputs

Name	Type	Description
TransformationFunction instance	TransformationFunction	Callable that transforms data points
__call__ result	Optional[DataPoint]	Modified data point or None if not applicable

Usage Examples

import random
from snorkel.augmentation import transformation_function

@transformation_function()
def tf_replace_word(x):
    """Replace a random word with a synonym."""
    words = x.text.split()
    if len(words) == 0:
        return None
    idx = random.randint(0, len(words) - 1)
    words[idx] = "REPLACED"
    x.text = " ".join(words)
    return x

@transformation_function()
def tf_add_prefix(x):
    """Add a common prefix to text."""
    x.text = "FYI: " + x.text
    return x

# Test
from types import SimpleNamespace
dp = SimpleNamespace(text="Hello world")
result = tf_replace_word(dp)
print(result.text)  # e.g., "REPLACED world"

Related Pages

Implements Principle

• Principle:Snorkel_team_Snorkel_Transformation_Function_Definition