Source code for textattack.attack_recipes.pwws_ren_2019



(Generating Natural Language Adversarial Examples through Probability Weighted Word Saliency)

from textattack import Attack
from textattack.constraints.pre_transformation import (
from textattack.goal_functions import UntargetedClassification
from textattack.search_methods import GreedyWordSwapWIR
from textattack.transformations import WordSwapWordNet

from .attack_recipe import AttackRecipe

[docs]class PWWSRen2019(AttackRecipe): """An implementation of Probability Weighted Word Saliency from "Generating Natural Langauge Adversarial Examples through Probability Weighted Word Saliency", Ren et al., 2019. Words are prioritized for a synonym-swap transformation based on a combination of their saliency score and maximum word-swap effectiveness. Note that this implementation does not include the Named Entity adversarial swap from the original paper, because it requires access to the full dataset and ground truth labels in advance. """
[docs] @staticmethod def build(model_wrapper): transformation = WordSwapWordNet() constraints = [RepeatModification(), StopwordModification()] goal_function = UntargetedClassification(model_wrapper) # search over words based on a combination of their saliency score, and how efficient the WordSwap transform is search_method = GreedyWordSwapWIR("weighted-saliency") return Attack(goal_function, constraints, transformation, search_method)