Implementation:Snorkel team Snorkel PandasSFApplier Apply

Knowledge Sources	Snorkel
Domains	Data_Slicing, Robustness
Last Updated	2026-02-14 20:00 GMT

Overview

Concrete tool for applying slicing functions to a Pandas DataFrame to produce a named slice matrix, provided by the Snorkel library.

Description

PandasSFApplier extends PandasLFApplier with _use_recarray = True, so that the output is a NumPy record array with named columns matching each slicing function. This is a thin wrapper (7 lines) that reuses the full LF applier infrastructure.

Usage

Import this class when you have defined slicing functions and need to generate the slice matrix for slice-aware training or evaluation.

Code Reference

Source Location

• Repository: snorkel
• File: snorkel/slicing/apply/core.py
• Lines: L13-20

Signature

class PandasSFApplier(PandasLFApplier):
    """SF applier for a Pandas DataFrame.
    Inherits PandasLFApplier with _use_recarray = True.
    """

    def apply(
        self,
        df: pd.DataFrame,
        progress_bar: bool = True,
        fault_tolerant: bool = False,
        return_meta: bool = False,
    ) -> np.recarray:
        """
        Apply slicing functions to DataFrame.

        Args:
            df: Input DataFrame.
            progress_bar: Show progress.
        Returns:
            np.recarray with named columns per SF, shape [n_examples, n_sfs].
        """

Import

from snorkel.slicing import PandasSFApplier

I/O Contract

Inputs

Name	Type	Required	Description
sfs	List[SlicingFunction]	Yes	Slicing functions (passed to constructor as lfs)
df	pd.DataFrame	Yes	DataFrame of data points

Outputs

Name	Type	Description
S	np.recarray	Slice matrix [n_examples, n_sfs] with named columns; values 0 or 1

Usage Examples

Apply Slicing Functions

import pandas as pd
from snorkel.slicing import PandasSFApplier, slicing_function

@slicing_function()
def sf_short(x):
    return 1 if len(x.text.split()) < 5 else 0

@slicing_function()
def sf_has_link(x):
    return 1 if "http" in x.text else 0

sfs = [sf_short, sf_has_link]
applier = PandasSFApplier(sfs)

df = pd.DataFrame({"text": ["hi", "check http://example.com", "a long message here today"]})
S = applier.apply(df)

print(S.dtype.names)  # ('sf_short', 'sf_has_link')
print(S["sf_short"])   # [1, 0, 0]
print(S["sf_has_link"]) # [0, 1, 0]

Related Pages

Implements Principle

• Principle:Snorkel_team_Snorkel_Slicing_Function_Application