wehub-resource-sync
35 lines
1.2 KiB
Python
35 lines
1.2 KiB
Python
from __future__ import annotations
|
|
|
|
from typing import Any
|
|
|
|
import numpy as np
|
|
import numpy.typing as npt
|
|
|
|
from ._masker import Masker
|
|
|
|
|
|
class Fixed(Masker):
|
|
"""This leaves the input unchanged during masking, and is used for things like scoring labels.
|
|
|
|
Sometimes there are inputs to the model that we do not want to explain, but rather we want to
|
|
consider them fixed. The primary example of this is when we explain the loss of the model using
|
|
the labels. These "true" labels are inputs to the function we are explaining, but we don't want
|
|
to attribute credit to them, instead we want to consider them fixed and assign all the credit to
|
|
the model's input features. This is where the Fixed masker can help, since we can apply it to the
|
|
label inputs.
|
|
"""
|
|
|
|
shape: tuple[None, int]
|
|
clustering: npt.NDArray[Any]
|
|
|
|
def __init__(self) -> None:
|
|
self.shape = (None, 0)
|
|
self.clustering = np.zeros((0, 4))
|
|
|
|
def __call__(self, mask: Any, x: Any) -> tuple[list[Any], ...]: # type: ignore[override]
|
|
return ([x],)
|
|
|
|
def mask_shapes(self, x: Any) -> list[tuple[int, ...]]:
|
|
"""The shape of the masks we expect."""
|
|
return [(0,)]
|